The workflow you already wrote is probably enough

Earlier this month, I went looking for a prompt workflow that I knew I had written. That’s usually a bad sign. The workflow already existed, the thinking was solid, and the structure still made sense, but I still had to find the post, copy the correct pieces, paste them into Claude, and move the outputs somewhere else.

That’s where most prompt libraries start to break down. The prompt usually holds up. The operating model around it breaks first, because a good workflow trapped in a post, Google Doc, or Notes still asks you to coordinate too much by hand.

I have more than 100 prompt workflows sitting inside the TinyTechGuides inventory. Some are useful references. A smaller number deserve promotion into something I can run without hunting through old posts. For this tutorial, I’m using one of the buying-committee workflows from the lead magnet project, Hidden Objection and Risk Surface Analysis. It already has the bones of a Skill because it defines the inputs, follows a repeatable process, and gives a PMM or sales team an output worth running more than once.

The goal here is not enterprise automation nirvana. I want to take one workflow that already works, package it so Claude can run it cleanly, and make the next run easier than the last one. The thinking is already done, so the job is to stop rebuilding it by hand every time.

The example: hidden objections

The Hidden Objection workflow is built for a problem that most PMMs recognize. Sales teams prepare for the objections that buyers say out loud, such as price, timing, integrations, or missing features. The deals that disappear into silence usually die somewhere else, when someone sees career risk, political risk, implementation risk, or budget risk and decides that doing nothing feels safer.

That makes it a strong workflow for this exercise. It is not a one-line prompt that asks Claude to brainstorm objections. The analysis starts with inputs, moves through six steps, and ends with a risk surface map that marketing and sales can use. It also asks for the same inputs that a Skill should collect, such as product, market, deal size, committee context, sales notes, and win-loss data.

The steps already behave like a process. It surfaces hidden objections by stakeholder role, maps the “no decision” incentive structure, looks for organizational risk triggers, and builds a mitigation plan. A final pass turns the analysis into content recommendations and a usable sales-facing deliverable.

What makes this workflow convertible

Not every prompt workflow deserves to become a Skill. Some prompts are better left as one-off thinking aids. A naming brainstorm or rough first-draft helper probably does not need a folder, a command name, and a maintained instruction file.

The workflows that deserve conversion have a few traits in common. You run them more than once, they ask for structured inputs, and their output needs to look consistent every time. They also carry enough judgment that you do not want each user improvising the process from scratch.

The Hidden Objection workflow passes that test. A slash command could be /hidden-objection-analysis. Its required inputs already live in Step 0, the reasoning path follows the six-step sequence, and the output can always include a risk surface analysis, content recommendations, and a short executive summary.

Expectations matter as much as those criteria. A first Skill does not need to connect to Salesforce, Gong, and every other system in the company on day one. The first version should do what the original workflow already did, with fewer chances to lose the thread.

Strip the workflow down to its operating parts

The easiest way to convert a workflow is to stop treating it as prose. An article explains the idea and gives the reader context. A Skill needs the operating parts.

In this workflow, the operating parts are easy to spot. Step 1 asks Claude to act as a senior B2B sales psychologist and buying committee analyst. It gives the business context, then Claude identifies the surface objection, hidden objection, root fear, behavioral signal, and trigger event for each stakeholder.

That is the old prompt structure hiding in plain sight. Role. Context. Task. Format. Tone. I still like that structure because it forces the hard thinking before the model starts generating, but it is not the finished product.

For a Skill, those pieces become reusable instruction blocks. The role gives the Skill its perspective, and the context becomes required inputs. The task becomes the process. Format becomes the output contract, while tone keeps the analysis direct instead of generic.

Each part has a direct translation:

- Role: who the Skill should act as

- Context: the inputs the Skill must collect before running

- Task: the ordered process the Skill should follow

- Format: the required output sections and tables

- Tone: the quality standard for the finished deliverable

That mapping is most of the conversion. Separate durable operating instructions from one-time article copy, then make those instructions easy for Claude to invoke.

Turn those parts into SKILL.md

Claude Code Skills are built around a SKILL.md file. Anthropic describes Skills as reusable instructions that Claude can load when relevant or when you invoke them directly with a slash command.^[1] The file does not need to be complicated. It needs to tell Claude when to use the Skill, what inputs it needs, what process to follow, and what output to produce.

For the Hidden Objection workflow, a finished SKILL.md looks like this:

description: Build a hidden-objection risk surface analysis for a B2B buying committee. Use when the user wants to diagnose why enterprise deals stall, go quiet, or end in no decision.

argument-hint: “[company] [product] [solution_category]”

---

## Required inputs

Before running, collect or infer:

- Company and product

- Solution category

- Target industry and target organization size

- Typical deal size

- Existing personas or committee map

- Sales observations, win-loss notes, or stalled-deal patterns

If any required input is missing, ask for it before producing the analysis.

## Process

1. Identify the likely buying committee roles.

2. Surface hidden objections by stakeholder role.

3. Map the no-decision incentive structure.

4. Identify urgency accelerators and risk amplifiers.

5. Build a risk mitigation plan.

6. Recommend content and messaging that preempts the highest-risk objections.

## Output format

Return:

1. Executive summary

2. Hidden-objection table by stakeholder

3. No-decision incentive analysis

4. Risk trigger map

5. Mitigation plan

6. Content and messaging recommendations

## Quality bar

Write like a senior B2B product marketing advisor. Be candid, specific, and grounded in buying-committee behavior. Do not produce generic persona language.

That snippet is the compressed operating system for the workflow. The original prompt doc can still live as a reference file, especially if you want Claude to load the full step-by-step version when needed. The main SKILL.md should stay concise because the loaded Skill text stays in context once the Skill runs.^[2] You can type that file out by hand, but it is faster to have Claude draft it from the workflow you already wrote, then check the draft against the shape above.

Have Claude draft the first version

Paste the prompt workflow into Claude and ask it to produce that structure for you:

I want to convert this prompt workflow into a Claude Skill.

Review the workflow below and create:

1. A concise `SKILL.md`

2. A clear description and argument hint

3. Required inputs

4. Core process steps

5. Expected output format

6. Quality rules

7. Any reference files that should sit beside the Skill instead of inside `SKILL.md`

8. A recommendation on whether this should be one Skill, several smaller Skills, or one orchestration Skill that points to companion Skills

Do not copy the workflow directly. Compress it into durable operating instructions.

Here is the workflow:

[PASTE WORKFLOW]

The first draft will probably be too literal, and that is fine. Ask Claude which parts are durable instructions, which details belong in a reference file, what inputs are required, where the Skill should ask clarifying questions, which steps can be combined, and what tends to break.

Atomization is the judgment call here. Do not split a workflow into separate Skills just because the original article had separate steps. Split it only when a subtask can stand alone and will be reused in other workflows. In the PMM stack, buying committee mapping, hidden-objection analysis, and committee-aware messaging could each become separate Skills over time, with a higher-level Skill telling Claude when to use those companion workflows and how to assemble the final deliverable.

Portability starts to matter once the workflow becomes part of the operating system. I am using Claude because it is where I run this marketing stack today. Codex and Gemini CLI now support agent skills built around SKILL.md, and both have project-context files that play a similar role to CLAUDE.md.^[3] The names change by tool, but the operating idea holds. Repeated instructions belong in reusable workflow files, not in a prompt you keep pasting from a browser tab.

What changes when it becomes a Skill

You feel the difference the first time you run it. A prompt workflow says, “Here is how to do the work.” A Skill says, “Run this job.” With the old workflow, you open the source post, copy the prompt, paste the inputs, run the step, move the output forward, and keep going until the final analysis comes together.

With the Skill, the orchestration moves into the instruction file. You type something like /hidden-objection-analysis “Acme Analytics” “Decision intelligence platform” “enterprise analytics” and Claude knows what job to run. It can ask for missing inputs and return the output in the agreed format.

Each manual step has a Skill equivalent:

- A prompt workflow step becomes a SKILL.md instruction

- Pasting variables by hand becomes a required-input list

- Copying outputs across steps becomes a Skill-managed process

- A final doc assembled by hand becomes a standard output format

- A one-off chat run becomes a reusable slash command

The Skill gets better when it reads the rest of the project context. CLAUDE.md can tell it the company positioning, voice rules, output conventions, and where source files live. Memory can preserve recurring preferences. MCP connectors can eventually pull sales notes or account context instead of asking you to paste everything into chat.^[4]

You do not need all of that on day one. The first useful version removes the copy-paste coordination. Later versions can read project context or reach into tools when the workflow deserves that extra plumbing.

The conversion pass I’d run first

If I were converting this workflow in a client project, I would start with one focused pass and stop when the Skill can produce a credible first output. The perfect version can wait.

1. Pick the slash command name. Use a verb phrase that describes the job, not the source document. /hidden-objection-analysis is better than /workflow-10 because it names the problem.

2. Pull the Step 0 variables into a required-input list. The workflow already has the correct inputs; they just live in article form instead of an input contract.

3. Compress the six workflow steps into process instructions. Keep the order, but cut the repeated setup. The Skill does not need six separate role blocks.

4. Lock the output format. For this workflow, I would require an executive summary, a stakeholder table, a no-decision analysis, a risk trigger map, and content recommendations. Readers trust the Skill when the shape stays stable.

5. Test it on one real stalled deal. A real opportunity exposes whether the Skill asks for the right inputs, notices the right stakeholder risks, and produces something sales would use.

That is the first 30 minutes. The next pass can add examples, reference files, scripts, or MCP access. Aim first for a repeatable instruction file that runs without you babysitting each step.

Promote the best workflows

Prompt workflows were the right unit for the first wave of AI marketing work. They forced marketers to define the role, context, task, format, and tone before asking the model to produce anything, and that discipline still matters.

The useful workflows now need a promotion path. Some can stay as posts, some belong in a gated PDF, and a smaller number should become Skills because they represent work you want to run repeatedly and improve over time. That last group is why the lead magnet matters. The PMM’s Prompt Playbook gives readers 11 workflows that solve real product marketing problems, and the next step is to choose the first one to convert, name the command, and make it easier to run the second time.

Start with the workflow that already gets reused, package it as a Skill, then run it once, fix what breaks, and let the next run inherit the improvement.

The prompt library was the raw material. The Skill is the operating procedure.

Share

Leave a comment

Frequently asked questions

What is a Claude Skill?

A Claude Skill is a reusable instruction folder that Claude Code can load when relevant or when you call it with a slash command. The core file is usually SKILL.md, which tells Claude when to use the Skill, what inputs it needs, what process to follow, and what output to produce. For marketers, a Skill turns a repeatable prompt workflow into something closer to an operating procedure, part of the broader move from chatbots toward agentic AI systems that plan work and act across tools.^[5]

How do I convert a prompt workflow into a Claude Skill?

Paste the existing workflow into Claude and ask it to draft a concise SKILL.md with a description, argument hint, required inputs, process steps, output format, quality rules, and supporting reference files. Claude will often copy too much from the source on the first pass, so your job is to compress it into durable instructions and test it on a real example.

Should every prompt workflow become a Skill?

No. Convert workflows that you run more than once, require structured inputs, produce a stable output, and carry judgment that should not be reinvented each time. A brainstorming prompt, summary helper, or rough first-draft prompt can stay in your library.

Should one workflow become one Skill or several Skills?

Start with one Skill unless a subtask can stand on its own and will be reused elsewhere. Splitting a six-step article into six Skills usually creates more overhead than value. In a PMM stack, buying committee mapping, hidden-objection analysis, and committee-aware messaging could each support other workflows, so a higher-level Skill can orchestrate them as companions.

Do Codex and Gemini CLI support a similar idea?

Yes. Codex and Gemini CLI both support agent skills built around SKILL.md, and both have project-context files that work like a project rulebook. The file names and implementation details differ, but the operating principle is the same. Repeated instructions belong in reusable workflow files that the agent can load when needed.

What should I ask Claude after it drafts the Skill?

Ask which instructions are durable, which details should move into a reference file, what inputs are required, where the Skill should ask clarifying questions, which steps can be combined, what tends to break, and how to test it on a real stalled deal. The first draft is a starting point, not the final asset.

About David Sweenor

David Sweenor is a Top 25 AI thought leader, author, and founder of TinyTechGuides. He spent the first half of his career as a practitioner at IBM in data science, business intelligence, and data warehousing. The second half he led product marketing teams at SAS, Dell Software, Quest, TIBCO, Alteryx, and Alation across advanced analytics, AI, and B2B marketing transformation. He writes about AI for marketers, Claude Skills, prompt workflows, and B2B operator depth at TinyTechGuides.

Books

- Artificial Intelligence: An Executive Guide

- Generative AI Business Applications

- The Generative AI Practitioner’s Guide

- The CIO’s Guide to Adopting Generative AI

- Modern B2B Marketing

- The PMM’s Prompt Playbook

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Anthropic. “Extend Claude with skills.” Claude Code Docs. Accessed June 16, 2026. https://docs.anthropic.com/en/docs/claude-code/skills.

^[2]Anthropic. “Extend Claude with skills.” Claude Code Docs. Accessed June 16, 2026. https://docs.anthropic.com/en/docs/claude-code/skills.

^[3]OpenAI. “Agent Skills.” Codex Manual. Accessed June 16, 2026. https://developers.openai.com/codex/codex-manual.md. Google. “Agent Skills.” Gemini CLI Docs. Accessed June 16, 2026. https://geminicli.com/docs/cli/skills/. Google. “Provide context with GEMINI.md files.” Gemini CLI Docs. Accessed June 16, 2026. https://geminicli.com/docs/cli/gemini-md/.

^[4]Anthropic. “Connect Claude Code to tools via MCP.” Claude Code Docs. Accessed June 16, 2026. https://docs.anthropic.com/en/docs/claude-code/mcp.

^[5]Purdy, Mark. “What Is Agentic AI, and How Will It Change Work?” Harvard Business Review, December 12, 2024. https://hbr.org/2024/12/what-is-agentic-ai-and-how-will-it-change-work.

I spent the first half of my career at IBM, where I built dashboards, predictive analytics solutions, and complained about our data warehouse. Then, I joined the EDW team to try to fix it, ran an analytics development team, and eventually landed in their analytics center of excellence (CoE). Similarly, Andreas Welsch spent close to twenty years at SAP, finishing as the VP who ran their AI Center of Excellence. We both left to run our own businesses and learned similar lessons. When you’re an independent entrepreneur, you become the CXO of everything, from revenue to legal to accounting to the marketing that nobody else is going to do for you.

That shared vantage point made our conversation on the Data Faces Podcast easy from the start. Andreas has watched hype-infused trends play out four times now: cloud, mobile, the first wave of machine learning, and now generative and agentic AI. Each time, the patterns that follow are similar.

“We have this new shiny object. Let’s go figure out what we can do with this. Throw spaghetti at the wall and see what sticks.”

— Andreas Welsch, Founder and Chief Human Agentic AI Officer, Intelligence Briefing

When you’re chasing the latest thing, sometimes the spaghetti sticks to the wall, while other times the house of cards comes crashing down. The companies that come out ahead, Andreas argues, are the ones that stop to ask a question most leaders skip under this much pressure. Just because you can build something with AI does not mean you should.

About Andreas Welsch

Andreas Welsch is the founder and Chief Human Agentic AI Officer at Intelligence Briefing, where he helps business leaders figure out what to do with AI. He spent close to two decades at SAP, finishing as the vice president who ran the company’s AI Center of Excellence, so he watched enterprise AI grow up from the inside. He is the author of two books, The AI Leadership Handbook and The Human Agentic AI Edge, an adjunct professor in Pennsylvania, a LinkedIn Top Voice, and the host of the What’s the BUZZ? podcast. The engineering curiosity started early. There are photos of him around four or five years old, screwdriver in hand, taking apart an RC car to see how it worked, then ending up with a small pile of leftover springs and screws.

In our conversation, Andreas and I covered:

- Why the rush to cut headcount with AI spreads like a contagion, and the revenue question almost nobody asks

- The “should we?” test that separates real value from sunk cost

- What the “SaaS is dead” crowd gets wrong about convenience, risk, and who you call at 2 a.m.

- How he used three custom GPTs to edit his book, and where AI’s help turned into noise

- Why agentic AI risk multiplies rather than adds up as you stack more agents

The race nobody’s questioning

Before we got to agents, we talked about layoffs, because Andreas sees the two as interrelated. One company announces it needs fewer people and more technology, whether or not it has figured out how. The media picks it up, investors ask the competitor down the street why it is not running as lean, and like dominoes, the next company follows, until a single press release hardens into an industry expectation.

“Having layers that continue until the morale improves isn’t really the way to success. And we know this, and leaders know this too. Yet this is happening because somebody over here said they’re doing it.”

— Andreas Welsch, Founder and Chief Human Agentic AI Officer, Intelligence Briefing

The same contagion now drives agentic AI. One company says it is building agents, true or not, and everyone else picks up the language. I told Andreas that I have not seen many agentic workflows in production. I see prototypes, pilots, and a lot of experimentation, but turning an agent loose on the real world is still rare, and plenty of those pilots stall long before they reach production.^[1] He agreed we are at least past the slide-deck arguments over whether to call it “AI agents” or “agentic AI,” yet most organizations are still deciding which use cases are worth the effort.

The question that saves you

When I asked Andreas what worries him about all this posturing, he started with a claim he hears all the time. SaaS is dead. Every other LinkedIn post now declares the death of software because anyone can build their own. So he tested it. After upgrading his Claude subscription to the hundred-dollar tier, he started rebuilding the tools he pays for. He cloned DocuSign over a weekend, the signing boxes wired to an email workflow that saved the file. He rebuilt his workshop live-polling app in a few days, then knocked out four or five more, from digital sticky notes to a credentialing tool.

The experiment worked, and that is exactly what taught him the lesson.

“You’re actually paying for convenience and for peace of mind when you get a SaaS subscription. There’s somebody else who is maintaining that thing for you. For 20 dollars a month? That’s actually a pretty good deal.”

— Andreas Welsch

A twenty-dollar subscription suddenly looks cheap when you remember what it covers. Someone else handles the dependencies, security patches, data-privacy rules, and the small stuff like getting the fonts to line up. Rebuilding a non-essential app for personal use is a fun exercise, but rebuilding the core systems a business runs on is a different calculation. ERP, CRM, and finance software need auditability, and when something breaks at two in the morning on a Sunday, you want a vendor on the hook to fix it, not a teammate who vibe-coded the thing last weekend. What does not change is the question under every build-or-buy decision. Just because you can build it does not mean it belongs on your plate.

Cost, or revenue?

Underneath the layoffs and the refactoring, Andreas keeps waiting to hear leaders ask one question. How are you going to make more money? He hears plenty about trimming costs and protecting margin. He rarely hears anyone ask where new revenue is supposed to come from.

“I wish there were more people asking, so how are you making more money? Not how are you optimizing your costs? Revenue is a lot harder to achieve, and building products that people want to buy and offering services that people need, it’s a lot harder to do than taking out costs.”

— Andreas Welsch

This is where his optimism diverges from how most companies behave. The same technology leaders use to justify cuts could instead help a team do ten times more, build new products, and reach customers it could not serve before, without sacrificing the people who would create that growth. Most want the incremental win with a smaller headcount, and the people who stay do the work of five.

The market data backs up his skepticism about where the value lands. McKinsey’s 2025 State of AI survey found that only about 39 percent of organizations report any measurable effect on enterprise earnings from AI, and most of those credit it with less than five percent. Sixty-two percent say they are at least experimenting with AI agents, yet only 23 percent are scaling them.^[2] The enthusiasm shows up everywhere. The financial return, for most companies, has not arrived yet.

What AI still gets wrong

The clearest picture of where AI helps and where it stops came from Andreas’s own book. He had planned to hire a human copy editor and line editor, the way he had before, because he likes the coaching and back-and-forth. Friends pushed him to let AI do it instead. So he wrote the manuscript himself with no AI, then built three custom GPTs, one a developmental editor, one a copy editor, and another a line editor, and fed his draft through all three.

The results were promising. The AI caught inconsistencies and even factual errors, things he had misremembered from news stories that a human editor would likely have missed. Then it kept going.

“AI, or in this case ChatGPT, was a helpful assistant that really didn’t know when to shut up.”

— Andreas Welsch

Every new revision came back with another five urgent fixes, then five more, until the suggestions started making the book worse instead of better. He was watching diminishing returns in real time, and he realized the skill he needed was knowing enough about his own craft to say “this far, and no further.” Without that line, you cannot tell whether the system is improving your work or making it worse.

That same limit scales up to the autonomous-enterprise vision everyone keeps selling. I asked Andreas which piece of conventional wisdom about agentic AI he thinks is most wrong, and he did not hesitate.

“It does everything for you, and it does it perfectly all the time. We’re still relying on a probabilistic system that can be confidently wrong.”

— Andreas Welsch

Even with governance, guardrails, and evaluation in place, an agent still has enough room to do something nobody wanted. And the risk does not add up the way people assume. One agent is manageable, two working together get more complex, and by the time you are orchestrating several, the risk compounds exponentially. Most companies are still building their first or second one while the industry sells them autonomous enterprises. Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027, undone by rising costs, unclear business value, and weak risk controls.^[3] Plenty of that failure traces back to strategy and governance rather than the models themselves, the same conclusion behind the forecast that most generative AI projects will fall short of their goals.^[4]

The human edge

Andreas gave himself a title that sounds like a contradiction, Chief Human Agentic AI Officer, and by the end of our conversation, it made sense. The leaders getting real value from AI share a habit. They treat the technology as a way to expand what their teams can do, keeping a human in the loop to decide what is worth doing at all. The most useful AI deployments I see amplify human judgment instead of replacing it.^[5]

Human judgment is the whole game. A manager asks how the company will make more money before reaching for another round of cuts, and an author learns when to stop taking the model’s notes. The same instinct tells an executive to pause before automating a process just because a vendor swears it can be done. Agentic AI will keep getting more capable, and the pull to hand everything over to it will keep getting stronger. The advantage goes to the people who can look at all that capability and still ask the oldest question in business. Should we?

Listen to the full conversation with Andreas Welsch on the Data Faces Podcast.

Based on insights from Andreas Welsch, Founder and Chief Human Agentic AI Officer at Intelligence Briefing, featured on the Data Faces Podcast.

Share

Leave a comment

Podcast highlights

- [1:01] What Intelligence Briefing does, and helping leaders decide what to do with AI

- [1:48] From wanting to be a pediatrician to taking apart RC cars with a screwdriver

- [3:45] Leaving SAP and becoming the CXO of everything

- [9:59] The optimism gap, and why so many teams are burned out doing five jobs

- [10:40] The layoffs vicious cycle, and the revenue question nobody asks

- [13:58] Pilots versus production, and why people have gone quiet about what they are building

- [21:37] “SaaS is dead,” and vibe-coding clones of DocuSign and Mentimeter

- [25:00] When to defer risk to a vendor, and the shift away from per-seat pricing

- [27:57] Just because you can does not mean you should

- [32:47] Editing a book with three custom GPTs that would not stop talking

- [36:12] The conventional wisdom he thinks is wrong, and why agent risk compounds

Frequently asked questions

What is agentic AI, and how is it different from a chatbot or generative AI?

Generative AI produces content such as text, code, or images in response to a prompt. Agentic AI goes a step further by taking actions, connecting to other tools, and completing multi-step tasks with some degree of autonomy. In the episode, Andreas Welsch describes agents that can run parts of a workflow on their own. The catch is reliability. Because the underlying system is probabilistic, an agent can act confidently and still be wrong, which is why human oversight matters.

Why do most AI agent projects fail to reach production?

Most agent efforts stall because organizations chase the technology before defining the value. Gartner predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027 because of rising costs, unclear business value, and weak risk controls. Companies that succeed start with the workflow they want to change, decide what to eliminate, and measure value from the beginning rather than launching a pilot and hoping it finds a purpose.

Does using AI mean cutting headcount?

It does not have to. Andreas Welsch argues that the bigger opportunity is using AI to help existing teams do far more, build new products, and reach new customers, which protects future growth rather than trading it away for a short-term cost cut. McKinsey’s 2025 research found that only about 39 percent of organizations report any measurable earnings impact from AI so far, a sign that headcount cuts alone do not deliver the promised return.

Should a company build its own software instead of paying for SaaS?

It depends on whether the software is core to the business. Andreas Welsch rebuilt several non-essential personal tools to prove it was possible, then concluded that a subscription often pays for convenience and peace of mind. Someone else handles maintenance, security, and data privacy. For core systems like ERP, CRM, or finance, auditability and vendor support usually outweigh the savings from building it yourself.

Where should a leader start with agentic AI?

Start with a single high-value workflow rather than a broad rollout. Ask whether the project should be done at all, not just whether it can be. Keep a human in the loop to judge quality, and treat reliability and risk as first-order concerns because agent risk compounds as you add more agents. The goal is measurable value on one process before scaling to the next.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Herrera, Catalina. “Your Netflix Moment: Why CIOs Must Act Now on AI Agents (or Risk Becoming the Next Blockbuster).” TinyTechGuides Insights, October 7, 2025. https://insights.tinytechguides.com/p/your-netflix-moment-why-cios-must.

^[2]McKinsey & Company. “The State of AI in 2025: Agents, Innovation, and Transformation.” QuantumBlack, November 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-his craft well enough to say,state-of-ai.

^[3]Gartner. “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027.” June 25, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027.

^[4]Carlsson, Kjell. “AI in 2025: Why 90% of Gen AI Projects Will Fail.” TinyTechGuides Insights, March 22, 2025. https://insights.tinytechguides.com/p/ai-in-2025-why-90-of-gen-ai-projects.

^[5]Magne, Matt. “Augmented Intelligence: The Future of Sales Enablement.” TinyTechGuides Insights, November 4, 2025. https://insights.tinytechguides.com/p/augmented-intelligence-the-future.

In late May, I spent three days at the BARC Data and Analytics Retreat at Devil’s Thumb Ranch in Colorado. It’s a unique event. Instead of a big stage and a passive audience, you get a room of twenty-five or thirty data and AI leaders who interrupt each other, disagree out loud, and keep the conversation honest. Five minutes into the first presentation, someone already said, “I don’t agree with that,” and that set the tone for the whole retreat.

I wasn’t the only one who noticed. “You’re with a group of twenty or thirty people who have been in this vertical for a long time, and the discussion opens up in both directions,” Shree Neve of ClicData told me. John Colthart of Una AI pointed at the mix of people in the room. “When you look at the people here, the different types of businesses, it gives you such a rich context arena to throw around ideas. That level of diversity of opinion and thought, that part’s really cool.” And Ben Schein of Domo named something you rarely see at a vendor event, competitors trading notes in good faith. “Some of these people we’re competing with on deals and for customers, but it’s nice to come together and learn in a way that’s not giving away any secrets.”

Between sessions, I pulled a handful of people aside for short on-location conversations for the Data Faces Podcast. The topics ranged from data sovereignty to financial planning to the future of business intelligence, but one theme kept surfacing across every conversation. What decides whether AI delivers value right now is governance and control, plus context and a clear point of view about what you are building, far more than any model feature.

So I did what you do after a few days on a Colorado ranch. I rounded up the six conversations that stuck with me. Here they are.

Carsten Bange on why sovereignty is really about control

Dr. Carsten Bange, founder and CEO of BARC, gave one of the retreat’s first sessions, a talk on data sovereignty. The term gets used loosely, so he separated it into three nested ideas. Data sovereignty sits inside digital sovereignty, which also covers processes and technology, and AI sovereignty is the newer layer focused on who controls the models you run. BARC’s own Data Sovereignty 2026 survey found that 89% of organizations now call sovereignty important, with “very important” climbing from 42% to 51% in a single year, and US political developments jumping to a top-three driver at 54%.^[1] The headline that surprised even Carsten was geographic. US companies rate sovereignty as more important than European ones, and they are investing more in it, even though Europe wrote most of the rules.

Shawn Rogers on innovation outpacing governance

Shawn Rogers, CEO of BARC US, has a blunt read on where most companies sit with AI governance. He described a talk where he asked a few hundred people whether they had launched an AI agent, and every hand went up. When he asked who felt comfortable with how they govern it, almost every hand dropped. He puts roughly 20% of the organizations he talks to in the category of having real governance in place, which leaves the other 80% moving fast and hoping nothing breaks. He also walked through the financial side that catches teams off guard, the surprise bills that land on Monday morning after someone launches an agent on Friday afternoon, including one company that handed Claude to 12,000 employees with no budget at all.

Ben Schein on the question most teams skip

Ben Schein, Chief AI and Analytics Officer at Domo, framed the smartest filter for any AI decision around whether you should act, even when you can. With a general-purpose model, almost anything is technically possible, so the harder and more useful question is whether you should once you weigh governance, cost, and risk. He also reframed sovereignty in a way that stuck with the room, describing it as control and visibility over your data and what it is doing, rather than a question of where the data center physically sits. Token cost ran underneath the whole conversation, shaping which AI projects are worth running and which ones burn budget for little return.

Shree Neve on confidence outrunning capability

Shree Neve, VP of Operations at ClicData, gave the sharpest warning of the retreat for anyone rushing to bolt AI onto their data. Bad inputs do not produce obviously bad outputs. They produce confident, well-formatted, completely wrong answers, and you might not catch the problem until it has already shaped a decision. That same disconnect between confidence and capability showed up in BARC’s research too. In the Unstructured Data for AI study, 71% of leaders said they were confident they could extract value from their data, yet one in three admitted to lineage and control gaps, and data quality has now climbed to the single most cited measure of AI success at 48%.^[2] She pushes a refreshingly old-fashioned sequence. Start with the business decision you want to make, work backward to the question, and only then go find the data and the tool.

John Colthart on the number finance forgets

John Colthart, Chief Product Officer at Una AI, came into financial planning with a contrarian view after a long career across sales, marketing, and product. Most planning tools obsess over controlling spend, and in doing so they ignore the number that tells you whether the business is growing. He also refuses to force a false choice between Excel, a web portal, and AI, since most companies still run real planning in spreadsheets and probably always will. His foundation-first instinct matches what BARC sees across the office of finance. Data management ranks as the top corporate performance management priority at 8.2 out of 10, while generative AI for planning sits near the bottom at 4.6, and only 6% of organizations have AI in active production for performance management.^[3]

Ivan Vakhmyanin on building for trust

Ivan Vakhmyanin, co-founder of Visiology, is doing the uncomfortable thing on purpose. Ten years into building a business intelligence company, he is rebuilding the product from scratch as an AI-first system rather than bolting assistants onto the old one. His reasoning is direct. If he does not disrupt his own product, a competitor eventually will. The harder engineering choice underneath that is trust. He kept Visiology’s tested data engine and methodology on the back end, gave users a familiar chat-style experience on the front, and made every step traceable so people can verify how the system reached an answer. That instinct lines up with what Kevin Petrie called “vibe slop” in his retreat session, the failure that happens when teams deploy agents on shaky foundations without the governance and context to back them up.

Rounding up what I heard

Six conversations, six corners of the industry, and the same idea underneath each one. Carsten framed sovereignty as control. Shawn weighed how fast companies launch AI against how slowly they govern it. Ben asked whether you should, even when you can. Shree showed how confidence outruns capability when the data is weak. John argued for the foundation before the AI layer. Ivan engineered for trust so people can rely on what the system tells them. None of them led with model features, because features are no longer where the value or the risk lives. What separates teams getting real returns from teams burning budget is governance and control, plus the context and point of view behind what you build.

If you want the full conversations, all six are on the Data Faces Podcast. New episodes drop every couple of weeks, and the on-location interviews like these land between the studio conversations.

If you’d like to learn more about BARC, its research, and the retreat, visit barc.com.

Share

Leave a comment

Frequently asked questions

What is the BARC Data and Analytics Retreat?

The BARC Data and Analytics Retreat is an invitation-only event hosted by the analyst firm BARC, held in 2026 at Devil’s Thumb Ranch in Colorado. It gathers a small group of data, analytics, and AI vendor leaders for working sessions and open debate rather than stage presentations. The intimate format encourages the kind of disagreement and discussion that larger conferences rarely produce.

What is the difference between data sovereignty, digital sovereignty, and AI sovereignty?

According to Carsten Bange of BARC, the three are nested. Data sovereignty is part of digital sovereignty, which is broader and also covers processes and technology. AI sovereignty is a newer layer that focuses on AI-specific questions, most importantly who controls the models an organization uses. All three center on control and visibility rather than only the physical location of data.

Why is AI governance such a concern right now?

Adoption is moving faster than control. Leaders at the retreat described companies launching AI agents widely while only a minority have real governance over the underlying data, models, and agents. The result is uncontrolled risk and surprise costs, including organizations that gave large groups of employees access to AI tools with no budget or oversight in place.

Where should a team start with AI on their data?

Start with the business decision you want to make, not the tool. As Shree Neve of ClicData put it, work backward from the question to the data and only then choose the AI tool. Feeding AI weak data produces confident but wrong answers, so fixing data quality and governance first matters more than picking a model.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]BARC. “Data Sovereignty 2026: Reality, Relevance, Roadmap.” BARC, 2026.

https://barc.com/

^[2]Adrian, Merv, and Kevin Petrie. “Harnessing Unstructured Data for AI Innovation.” BARC Research Study, 2026.

https://barc.com/

^[3]BARC. “CPM Trend Monitor 2026 / The Planning Survey 26.” BARC, 2026.

https://barc.com/

Twenty-five years, same fight

Over the past 25 years, I’ve seen my share of hype cycles. Sometimes it feels like Groundhog Day, with things on repeat. At IBM, I built predictive analytics solutions and data warehouses. At SAS, Dell, TIBCO, and Alteryx, I marketed advanced analytics to companies that said they wanted to be data-driven, only to watch them revert to their complex spreadsheets the next day. Every five to seven years, a new shiny technology shows up that promises to usurp the previous one, and every time, the work is the same. You clean up your data, you get leadership aligned, and you convince people to change how they make decisions.

The current wave is agentic AI, and the technology behind it is certainly impressive. Frontier models can write better code than most engineers, pass the bar exam without breaking a sweat, and reason through problems that used to require a PhD. Anthropic, OpenAI, and the rest of the foundation model crowd are racing toward something that looks an awful lot like AGI. Meanwhile, on the 101 corridor through San Francisco, every billboard is selling agents, and every shoe company is now an AI company. Allbirds just signed a $50 million convertible facility to pivot into GPU-as-a-Service and rename itself NewBird AI, and the stock popped more than 350 percent on the announcement.

When I sat down with Josh Howard for episode 40 of the Data Faces Podcast, the topic he proposed was tongue-in-cheek on the surface, but beneath the surface lay a partial truth that most enterprises are still avoiding. Josh is the Senior Director of Product Marketing for Executive Audiences at Databricks. He and I first met at Dell more than a decade ago, then crossed paths again at Alteryx, where we were both trying to convince financial analysts that there was a better way than spreadsheets. His topic for the show was three words. Your AI is dumb. As Josh explained, the models themselves are some of the most advanced technologies that we have seen in our lifetime. However, they are only as smart as the data you give them, and most companies still haven’t figured out how to give them access to the data that matters most.

“Without context, your agents are dumb.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

About Josh Howard

Josh Howard is the Senior Director of Product Marketing for Executive Audiences at Databricks, where he has spent the last four years translating data and AI strategy for the C-suite. Before Databricks, we crossed paths in product marketing twice, first at Dell Technologies and then at Alteryx, where we spent our days trying to convince financial analysts that there was a better way than the spreadsheet. Outside of work, Josh lives in Colorado, ties his own fly-fishing lures, and told me on the show that if he weren’t doing product marketing, he would be a full-time fly-fishing guide on the rivers near Denver.

In our conversation on the Data Faces Podcast, Josh and I get into:

- Why “your AI is dumb” without enterprise context

- New findings from the Databricks and Economist Enterprise Making AI Deliver survey of 1,221 senior technology leaders, including the 84/43 measurement problem and why infrastructure costs more than the GPU bill

- Where agents are already changing how work gets done, and where they haven’t yet

- The cautionary tale of an agent who whacked a production database

- Josh’s contrarian take on the AGI debate

Why your AI is dumb without your data

AI has an insatiable appetite, but the models are often hankering for the wrong datasets. They were trained on the public internet, which makes them competent at history, cheating on homework, and bar exam questions. But, they have never seen your customer record, your forecast methodology, or the customer call recordings in Gong.

“These models have been trained on the internet. They’re really good at history or helping your kid do their homework. From an enterprise perspective, a lot of that work hasn’t been done to give it access to the data in your organization. You’ve got to have that context.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The data that your enterprise runs on is scattered throughout your organization. It sits in the systems where your customer relationships, your financial close, and your product telemetry live. Most of it is proprietary, much of it is unstructured, and the model has never seen any of it. Until it has access to that information, no amount of fine-tuning will make the answer any better.

This is the metadata fight from twenty years ago with a new name. Enterprise architects have been screaming about governance and consistent business definitions for two decades, and almost nobody on the business side was listening. Now those same arguments are showing up in CEO town halls because gen AI outputs have made the problem visible. Gartner has been making the same point, warning that organizations without AI-ready data will see most of their AI projects stall or fail through 2027.^[1] The product that Josh pointed at on the show is a conversational analytics layer trained on his organization’s internal semantics, policies, and nomenclature. A user types a question in plain English, and the system answers using the company’s own data, terminology, and rules. When your AI fails on a business question, the issue is rarely the model. It is almost always a data context problem.^[2]

The real cost isn’t the GPU bill

Talk to a CFO right now about AI, and the first word out of their mouth will be cost. The conversation will go straight to GPU pricing, vendor lock-in, and whether the AI bill will break the bank next quarter. Those are the visible costs. According to the new Databricks and Economist Enterprise Making AI Deliver survey of 1,221 senior technology leaders, the damage is happening somewhere else.^[3]

“Everyone is obsessing over the model cost and GPU spend, but the real tax there is actually the infrastructure underneath.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The survey asked leaders to identify their biggest AI cost concerns. Fifty-nine percent named data storage, movement, and duplication. Only 25 percent named compute. What the press and the boardroom focus on draws less than half the concern of the thing nobody talks about. The real cost is dragging your data from where it lives now to wherever the model needs it, and then doing it again three more times for the next system.

The payoff for fixing this is measurable. The same survey found that 97 percent of organizations with a unified data architecture report their AI investments are paying back ahead of plan.^[4][5] Almost nobody has a unified architecture today. Most enterprises run a hodgepodge of warehouses, application databases, and SaaS exports stitched together with batch jobs and prayer. The companies that have done that consolidation work are seeing returns. Everyone else is paying the tax twice.

The 84/43 problem

For most of my career, the architects, warehouse managers, and data scientists who understood the systems were screaming about governance, lineage, and consistent business definitions, and the people writing the checks weren’t listening. Then ChatGPT launched in November 2022. Almost overnight, the C-suite cared. Josh and I were both watching from inside product marketing, and the light bulb finally went off.

That attention brought real budget, executive air cover, and top-down sponsorship. Four years in, the Making AI Deliver survey shows where the bill is coming due. Eighty-four percent of senior executives say their AI returns are beating expectations, but only 43 percent require teams to measure the impact of those projects.^[6] Doesn’t that seem weird? Confidence has gotten well ahead of measurement, and the boardroom will eventually notice.

We’ve seen this pattern before. CRM in the late 1990s and big data in the early 2010s both produced euphoria first, then a wave of post-mortems and write-downs once boards started asking what the investment had returned. The 84/43 split is the present-day version of the same trap. Confidence without measurement holds up right until somebody in the boardroom asks for proof. When the proof comes, most AI projects don’t survive the audit.^[7]

This problem has a boring fix. Before any AI project starts, name the outcome that it should deliver, the metric that you will use to track it, and the executive who owns that metric. This isn’t rocket science; in fact, it’s the same advice that Gartner has been giving for 20 years. Then put a calendar reminder six months out so somebody opens the dashboard. That is the entire intervention. The companies on the right side of the next post-mortem are the ones doing this work today.

“There was a big paradigm shift with ChatGPT in November of 2022, where the light bulb really went off in the C-suite.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The engineering exception

The strongest place where agents are already working is in software engineering. Databricks publishes its own platform data on this. Two years ago, AI agents created 0.1 percent of databases on the Neon serverless Postgres layer. By October 2025, that number was 80 percent, with test and development environments climbing to 97 percent.^[8] The engineers building on top of Databricks are not writing database code by hand. They are reviewing what agents have shipped.

“Engineers aren’t banging away on the keyboard. They’re actually managing a team of agents.”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

Engineering worked first because the feedback is unambiguous. Code either compiles or it doesn’t, and decades of CI/CD automation have given agents a runway. Even so, the agents work under supervision. Last summer, a Replit coding agent deleted a SaaStr founder’s production database during a stated code freeze, despite explicit instructions to do no harm.^[9] Months of work disappeared in minutes. Human-in-the-loop is the price of admission for putting agents near production data.

The departments that most executives want to disrupt next (HR, sales, and marketing) do not look anything like engineering. The work is fuzzy, outcomes are negotiated, and unwritten rules carry as much weight as policy. Agents will get there eventually, but the path will be measured in years rather than quarters. The change management problems that Josh and I have spent careers writing about will matter more than the model capabilities.

The real race

Toward the end of our conversation, I asked Josh what will look obvious in 2027 that nobody believes today. His answer ran counter to the entire AGI news cycle. Josh argues that the next two years will not be about reaching superintelligence. For practical purposes, that race is already over. The model labs will keep pushing the capability frontier, and the headlines will keep getting louder. None of that will be where the money is made.

“The real race isn’t to superintelligence. Can you make the AI you already have actually work inside your company?”

— Josh Howard, Senior Director, Product Marketing for Executive Audiences, Databricks

The companies that will win the next five years will look boring from the outside. They will be the ones cleaning up their data, getting their semantics right, and tying every agent project back to the outcomes that leaders promised at the start. Boring work wins. AGI can wait… unless it’s already here.

Listen to the full conversation with Josh Howard on the Data Faces Podcast.

Based on insights from Josh Howard, Senior Director, Product Marketing for Executive Audiences at Databricks, featured on the Data Faces Podcast.

Share

Leave a comment

Frequently asked questions

What does it mean to say “your AI is dumb”?

The phrase comes from Josh Howard, Senior Director of Product Marketing at Databricks. Today’s frontier models are among the most advanced technologies ever built, and their training data comes from the public internet. They are excellent at history homework and bar exam questions, but cannot answer questions about your customer records, your forecast methodology, or your sales policies. Without access to that internal data, even the best model is dumb in the way that matters for your business.

Why is data infrastructure a bigger AI cost than GPUs?

According to the Databricks and Economist Enterprise Making AI Deliver survey of 1,221 senior technology leaders, 59 percent named data storage, movement, and duplication as their biggest AI cost concern, while only 25 percent named compute as their biggest AI cost concern. GPU spending is the visible bill. Most of the cost goes to transferring data between systems whenever a new AI application needs it. Organizations with a unified data architecture report AI investments paying back faster than those still stitching warehouses and SaaS exports together by hand.

Where are AI agents working in enterprises today?

The strongest evidence comes from software engineering. Databricks reports that AI agents now create 80 percent of new databases on its Neon serverless Postgres layer, up from 0.1 percent in 2023. Test and development environments climbed to 97 percent over the same window. The work is unambiguous, the feedback is fast, and decades of CI/CD automation have given agents runway. Other functions do not look anything like engineering, and the path for putting agents into HR, sales, and marketing will be measured in years.

How should I measure whether my AI investment is working?

Most organizations are not measuring it well. The Making AI Deliver survey found that 84 percent of senior executives believe their AI returns are beating expectations, but only 43 percent require teams to measure the impact. Confidence has gotten ahead of measurement. Fixing this is straightforward but unglamorous. Before any AI project starts, name the outcome it should deliver, the metric you will use to track it, and the executive who owns that metric. Then put a calendar reminder six months out to check the dashboard.

When will AI agents work for non-engineering functions?

Plan for a multi-year transition. AI agents already generate the majority of new database creations at companies like Databricks, but engineering has several advantages that other departments lack. The feedback is unambiguous, the outcomes are binary, and decades of CI/CD automation have given the agents runway. HR, sales, and marketing work is fuzzy, outcomes are negotiated, and culture and unwritten rules carry as much weight as policy. Change management problems will matter more than model capabilities.

Should I worry about AGI or focus on my company’s data?

Both, but only one is in your control. The frontier model labs will keep pushing toward something that looks like artificial general intelligence, and the headlines will keep getting louder. Your business does not get a return on those headlines. Your return comes from feeding agents the data, semantics, and policies that govern how your company makes decisions. The companies that will win the next five years will look boring from the outside, quietly consolidating their data and measuring outcomes while the press celebrates the latest billboard.

Podcast highlights

Timestamps estimated from the transcript and should be verified against the final cut.

[0:00] Opening and introduction

[1:17] Josh’s role leading PMM for executive audiences at Databricks

[2:21] If he weren’t doing PMM: full-time fly-fishing guide in Colorado

[3:23] “Your AI is dumb” — what the phrase actually means

[5:25] Structured vs. unstructured data and why the industry is still stuck in rows and columns

[6:20] Where Josh and Dave first met at Dell Technologies

[8:13] Metadata, context, and the 20-year-old enterprise architect fight

[9:37] The November 2022 ChatGPT moment when the light bulb went off in the C-suite

[11:07] Trying to pry Excel out of a financial analyst’s hands at Alteryx

[12:08] Human-in-the-loop and the Replit coding agent that wiped a production database

[12:53] Conversational analytics, Databricks Genie, and learning a company’s internal semantics

[19:11] Inside the Databricks and Economist Enterprise Making AI Deliver survey of 1,221 leaders

[20:54] The 84/43 measurement gap and why executive confidence is running ahead of proof

[23:21] The 59/25 cost split — data infrastructure costs more than GPUs

[28:30] Upskilling, the prompt engineer hype cycle, and why behavior change is the real bottleneck

[30:17] AI washing on the 101 corridor and Allbirds’ pivot to NewBird AI

[33:26] What will look obvious in 2027 — the real race isn’t superintelligence

[35:39] Closing thought: “Without context, your agents are dumb.”

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Gartner. “Lack of AI-Ready Data Puts AI Projects at Risk.” Gartner Newsroom, February 26, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk.

^[2]Sweenor, David. “Your AI Doesn’t Have a Model Problem. It Has a Data Context Problem.” TinyTechGuides, February 24, 2026. https://tinytechguides.com/blog/your-ai-doesnt-have-a-model-problem-it-has-a-data-context-problem/.

^[3]Economist Enterprise. “Making AI Deliver: A Benchmarking Framework on How Leading Companies Operationalise AI for Impact.” Sponsored by Databricks. 2026. https://www.databricks.com/resources/analyst-research/making-ai-deliver.

^[4]Economist Enterprise, “Making AI Deliver.” See note 1.

^[5]Sweenor, David. “Generative AI’s Force Multiplier: Your Data.” TinyTechGuides, October 14, 2023. https://tinytechguides.com/blog/generative-ais-force-multiplier-your-data/.

^[6]Economist Enterprise, “Making AI Deliver.” See note 1.

^[7]Sweenor, David. “How 3% of Companies Win with AI While 97% Fail.” TinyTechGuides, July 29, 2025. https://tinytechguides.com/blog/how-3-of-companies-win-with-ai-while-97-fail/.

^[8]Databricks. “2026 State of AI Agents: Enterprise Insights on Building AI.” 2026. https://www.databricks.com/resources/ebook/state-of-ai-agents.

^[9]Fortune. “AI-Powered Coding Tool Wiped Out a Software Company’s Database in ‘Catastrophic Failure.’” July 23, 2025. https://fortune.com/2025/07/23/ai-coding-tool-replit-wiped-database-called-it-a-catastrophic-failure/.

Three words defined the Qlik Connect 2026 keynote: context, trust, and freedom. Former Qlik CEO Mike Capone framed the stakes for enterprise AI on day one. It is not enough to produce a fluent answer. AI has to understand the business in context, run on a trusted foundation, and connect insight to action in the systems teams already use.^[1]

Capone’s larger thesis was that AI is moving from showcase to operating model.^[2] The way Qlik talks about users is part of that shift. In the back-to-back Data Faces conversations I did on the show floor with Mary Kern (VP Product Go-to-Market) and Brendan Grady (EVP and GM of Analytics & AI), old industry frames did not survive the table test. Mary said she was “never a fan” of “citizen data scientist,” and Brendan called the term “crazy” when it came up in his own Data Faces conversation.^[3] That is a signal worth paying attention to.

“I was never a fan of citizen data scientists for the record or citizen analyst.”

— Mary Kern, Vice President, Product Go-to-Market, Qlik

About Mary Kern

Mary Kern is Vice President, Product Go-to-Market at Qlik, where she leads marketing, launches, and product-led growth across the entire Qlik portfolio. She joined Qlik in 2023 leading product marketing for analytics and has since expanded her scope to cover the full product portfolio, including data integration, cloud, analytics, and AI. Before Qlik, she held marketing leadership roles at Varicent, SDL, TIBCO Software, and IBM, and holds an MBA from the Kellogg School of Management. Mary and I worked together at TIBCO years ago. When she isn’t shipping product keynotes she is running a suburban-Chicago wildlife cam and competing with me in an annual vegetable garden weigh-off.

In this episode, we discuss:

• Why “citizen data scientist” never worked as an industry frame

• How generative AI changes the question from “train users” to “meet users where they are”

• Designing for the user already in the seat, not the one we wish were there

• Where data quality and trust shift once natural language becomes the interface

• Qlik Connect 2026 themes and what practitioners should watch next

A design premise, not an enablement story

For 15 years, BI vendors pitched “citizen data scientist” as the answer for enabling non-data people. After endless debates about citizen dentists and citizen pilots, the term fizzled away. Most business users have no interest in becoming part-time data scientists. They want answers and recommendations on how to improve their business operations. Mary was direct about why. “It puts a lot of onus on people when that may not be their calling or aptitude,” she said.

Mary’s reply to that was a different design premise.

“Most people are horrible prompters. You have to bake that into the experience.”

— Mary Kern, Vice President, Product Go-to-Market, Qlik

Citizen data scientist asked the user to get better. With “horrible prompters,” the design question shifts to how the tool can get smarter about the user sitting in front of it. That reframes the work from enablement to design. For 15 years, self-service BI pushed the cognitive load onto the end user, who was expected to learn the data model, the query language, and the tool. Mary’s view is that generative AI changes the equation. It “really meets everybody where they’re at and their skill set.” Users don’t have to level up before getting an answer.

That design premise lines up with what Capone, Qlik’s former CEO, had been telling the market all year. Before the event, he described Qlik’s approach as helping teams engage data “through agentic conversations that lead to action, with governance and efficiency built in.”^[4] Mary’s design premise is the former CEO’s operating-model thesis at the UX layer.

Meeting users where they are

Mary’s design premise has a logical consequence. If the tool absorbs the skills burden at the UX layer, the quality and trust burden shifts one layer deeper.

“It really gets pushed down to one step behind analytics, which is the data product.”

— Mary Kern, Vice President, Product Go-to-Market, Qlik

With natural language as the interface, users stop worrying about field names or SQL syntax. The tool handles that. What users are actually depending on is the foundation underneath, whatever data the tool reaches for, and whether that data is correct.

Qlik’s Connect 2026 announcements make the shift concrete. Qlik Answers is the entry point, combining structured analytics and unstructured content. Discovery Agent surfaces anomalies before humans think to ask about them. Predict Agent builds models and answers forward-looking questions. Automate Agent pushes insights into downstream workflows. Analytics Agent accelerates development tasks.^[5] Together they form a continuous path from question to action.

Mary walked through what this looks like at the user level. The system takes a messy question and reframes it. It surfaces relevant data without requiring users to name fields or tables. When a question is ambiguous, the system flags it and asks for clarification instead of silently guessing.

Delivery matters as much as the pipeline. Rather than asking users to come to Qlik, the platform reaches users in whatever environment they already work in. MCP Server lets users invoke Qlik’s analytics engine from Claude, ChatGPT, Gemini, or whatever assistant their organization has standardized on. Brendan said Qlik is already seeing roughly 50/50 usage between its native interface and MCP for agentic capabilities. Agents run in the background, surfacing what matters without a dashboard login. The pane of glass is wherever the user already is.

That is “meeting users where they are” at the product level, not just the UX level. It is Capone’s “freedom” pillar executed in shipping code. For 15 years, the question was how to train more business users to work with data. Now the question is how to put trustworthy, governed data in front of users in the environments they already trust, including AI assistants that never ran inside the analytics stack.

Trust as a hard requirement

If the onus has shifted one layer deeper, that layer has to be trustworthy. Capone, who has since left Qlik, put it bluntly in the run-up to the event.

“AI is moving from an interesting capability to an operational expectation. The moment it touches real decisions, trust becomes a hard requirement, not a slogan.”

— Mike Capone, former CEO, Qlik

In an agentic era, the urgency is sharper. An agent doesn’t pause to gut-check a suspicious number. It takes the data, acts on it, and passes the result to the next step. By the time anyone notices a problem, the decision has already shipped.

Qlik’s response is to make trust operable. The Connect 2026 announcements on data products include a Trust Score that evaluates data products across accuracy, timeliness, diversity, and completeness. Data contracts define what a data product is expected to provide. The Data Product Agent helps teams create, manage, and evaluate data products using natural language.^[6] They turn trust into a visible operational signal rather than an assumed quality.

This reflects a deeper shift Capone signaled throughout his time leading Qlik. The old pendulum between tight central control and chaotic self-service is breaking down.^[7] What replaces it is controlled decentralization, with governed data products distributed to wherever users, human or agent, can make use of them. That requires governance, data contracts, semantic layers, lineage, and access controls to stop being back-office hygiene.^[8] They become the place where AI succeeds or fails.

Retire “citizen data scientist.” Invest in the data foundation and the delivery mechanisms that meet users in the environments they already work in.^[9]

Near the end of our interview, Mary captured the shift in one line. “We just have new ways of solving these old problems.” The hard part just stopped being the user’s job.

Listen to the full conversation with Mary Kern on the Data Faces Podcast.

Based on insights from Mary Kern, Vice President, Product Go-to-Market at Qlik, featured on the Data Faces Podcast.

Share

Leave a comment

Frequently asked questions

What does it mean to meet users where they are in product design?

Meeting users where they are is Mary Kern’s design philosophy for the agentic era. Instead of training users to phrase questions better, the tool absorbs the skills burden. Qlik’s agentic experience reframes messy questions, surfaces relevant data without requiring field names, and flags ambiguous questions instead of silently guessing.

What were the keynote themes at Qlik Connect 2026?

Former Qlik CEO Mike Capone framed the keynote around three words: context, trust, and freedom. AI has to understand the business in context, run on a trusted foundation, and connect insight to action in the systems teams already use. Qlik’s announcements at Connect 2026 extended this with MCP servers, agentic experiences, and an open ecosystem that meets users in whatever environment they already work in.

Where should D&A leaders invest to prepare for agentic AI?

D&A leaders should shift investment from end-user enablement programs to the data foundation underneath. Governance, data contracts, semantic layers, lineage, and access controls stop being back-office hygiene when AI becomes the interface. Agents don’t pause to gut-check suspicious data, so whatever they read has to be trustworthy before they touch it. That is where AI succeeds or fails.

How does natural language interface shift the burden away from users?

When natural language becomes the interface, users stop worrying about field names or query syntax. The tool handles that. What users actually depend on is the foundation underneath, whatever data the tool reaches for and whether it is correct. The burden moves from the user one layer deeper, to the data product layer that serves every question.

Podcast highlights

- [0:00] Introduction on the Qlik Connect 2026 show floor

- [0:24] Mary’s expanded role at Qlik

- [0:50] Gardens and a suburban raccoon cam

- [2:14] Qlik Connect 2026 keynote highlights

- [4:10] Qlik’s agentic experience and “a couple toggles to production”

- [6:30] What is different about enabling business users this time

- [8:20] Flexibility and meeting users where they are

- [11:15] The Qlik community

- [12:11] Cutting through the agentic noise

- [14:42] Storytelling and customer validation

- [16:13] What is next for Qlik in 2026

- [17:30] The “dare to be different” theme

About David Sweenor

David Sweenor is the founder and host of the Data Faces Podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Qlik. “Qlik Extends Analytics from Answers to Agentic Action.” Press release, April 14, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-extends-analytics-from-answers-to-agentic-action.

^[2]Qlik. “Qlik Connect 2026 Shows Enterprises Are Closer to Agentic AI Than They Think.” Press release, April 15, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-connect-2026-shows-enterprises-are-closer-to-agentic-ai-than-they-think.

^[3]Sweenor, David. “Why Bad Data Didn’t Matter Until Now.” TinyTechGuides, April 2026. https://tinytechguides.com/blog/why-bad-data-didnt-matter-until-now/.

^[4]Qlik. “Jesse Cole, Creator of the Savannah Bananas, to Keynote Qlik Connect 2026.” Press release, January 28, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/jesse-cole-creator-of-the-savannah-bananas-to-keynote-qlik-connect-2026.

^[5]Qlik. “Qlik Extends Analytics from Answers to Agentic Action.” Press release, April 14, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-extends-analytics-from-answers-to-agentic-action.

^[6]Qlik. “Qlik Makes Trust Operable for Data Products.” Press release, April 14, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-makes-trust-operable-for-data-products.

^[7]Qlik. “Qlik CEO: Enterprises Are Underachieving on AI, With Islands of Value in a Sea of Noise.” Press release, January 15, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-ceo-enterprises-are-underachieving-on-ai-with-islands-of-value-in-a-sea-of-noise.

^[8]Sweenor, David. “Why Bad AI Governance Kills 95% of Enterprise Projects Before Production.” TinyTechGuides, September 9, 2025. https://tinytechguides.com/blog/why-bad-ai-governance-kills-95-percent-enterprise-projects/.

^[9]Sweenor, David. “Why 80% of AI Projects Fail (And the Three Boring Decisions That Save the Other 20%).” TinyTechGuides, October 21, 2025. https://tinytechguides.com/blog/why-80-of-ai-projects-fail-and-the-three-boring-decisions-that-save-the-other-20/.

Last Tuesday, I was chatting with Claude Code and wanted to write a new B2B messaging prompt workflow.^[1] Then I vaguely recalled writing one last year, so I searched Substack and realized I had a couple of them sitting there since 2025, just waiting and wanting to be used again.

That caught me. When I’m going in circles with Claude Code, Gemini CLI, or Codex, my first inclination is to write yet another prompt workflow. Most marketers I know do the same thing, and that instinct was right a year ago. It isn’t anymore.

I’ve published more than 100 articles on AI, marketing, and prompt workflows on insights.tinytechguides.com since early 2025. Each one was useful the day it went live. Most of them are sitting in the inventory right now, ready for reuse, and many of them are doing more than that. They’ve formed the basis for the 60+ Claude Skills I use regularly today.

Two weeks ago, I wrote about the four-component stack that makes Claude reliable for marketing work.^[2] Skills do the work, and three supporting pieces, CLAUDE.md, memory, and MCPs, make the Skills compound. That post explained why each component, on its own, falls apart. This post answers the question I kept getting in the inbox afterward.

You don’t need more prompt workflows. You need to package, refine, and continually improve the ones you have. The conversion is faster than writing a new workflow from scratch, and it’s what the rest of this piece is about. On the continual improvement side, the /reflect skill and napkin.md are for a future post.

A prompt workflow is a one-shot, whereas a Skill compounds.

A prompt workflow is linear. You paste it into a chat window, fill in the variables, and run it.^[3] You copy the output somewhere useful and close the tab. Run the same workflow next month, and it forgets everything that you fixed last time. The model has no memory of which variant you settled on, which phrasings you cut, or which output format your client preferred. Was it a deck? A Google doc or a webpage?

Stack a hundred of those together, and you have a prompt library. A library is the right starting point. Each entry was useful the day that you wrote it, and each one is one trigger away from working again. Unfortunately, none of them know about each other’s capabilities, and the library essentially acts as a drawer full of sticky notes.

“A library of prompt workflows is dormant inventory. A library of Skills compounds every time you use it.”

— David Sweenor, Founder/CEO, TinyTechGuides

When most of us started writing prompt workflows, Claude Skills barely existed, and we were still enamored with ChatGPT. Everyone was talking about prompt libraries, prompt engineering, organizational context, and Custom GPTs. We wrote workflows, reused them, and shared Custom GPTs. That was the toolkit at the time. The tech has evolved, and the work hasn’t been lost. Every saved prompt workflow is raw material for a Skill.

A Skill is a different unit, a folder that Claude loads when you invoke it with a slash command. Loaded into a project that has its own CLAUDE.md, its own memory folder, and a few MCP connectors, the Skill stops being a recipe in a drawer. It becomes a recipe in a kitchen, with a smart pantry that knows your preferences and appliances wired into your accounts. The next time you run it, the kitchen remembers the corrections that you made the last time.

The next ten workflows I write won’t move the needle, but the ones I convert into skills will.

What a Skill adds that a prompt workflow can’t

A prompt workflow is a recipe. A Skill is a recipe plus three things a recipe alone never had.

The first is a trigger. A workflow lives in a doc somewhere. To run it, you have to find the doc, copy the prompts, paste them into a chat window, and orchestrate the steps yourself. Custom GPTs are a little better. You can fire one off without having to hunt for the doc. They’re a pain in the butt to update, though, because every tweak means logging back into the GPT builder to edit the instructions and hoping nothing else broke. A Skill solves that update problem because every correction lands in the conversation, not in a separate builder window.

A Skill is a slash command. Type /review-article and the workflow runs. The Skill is a verb you can say to the project, not a script you have to fetch.

The second is inheritance. A Skill reads the project’s CLAUDE.md and your memory folder every time it runs. The variables you filled in by hand were your product name, your client’s industry, and your usual output format. Now, they all come pre-filled from files that already exist. One Skill works across many projects without reconfiguration. Run the same /review-article Skill in two different client projects, and you get two different reviews of the same draft, both correct, because each project’s rulebook carries its own voice.

The third is tool access. A workflow that ends with “now go pull the latest win-loss notes from HubSpot or Salesforce and bring them back” asks you to do half the work. A Skill wired to a Model Context Protocol (MCP) server goes and gets the notes. Per Anthropic’s documentation, Skills are organized folders that load only when relevant, and they live next to the tools they need.^[4]

“A Skill is the recipe that remembers where the pantry is.”

— David Sweenor, Founder/CEO, TinyTechGuides

The second run of a workflow is identical to the first, except for a new set of variables. The second run of a Skill inherits everything that the project learned the first time, and the third run inherits everything from the first two. The work that you put in once does more work each time you call it.

From workflow to Skill

The Strategic Battlecard workflow is one of the more popular ones I’ve published at TinyTechGuides.^[5] It’s a 10-step prompt workflow that produces a single-page battle card for a single competitor. You enter the competitor’s name, your product, and your industry. Then you add the proof points the workflow asks for, such as win themes from recent deals, and Gong call analysis. It’s the kind of doc you reach for when sales complains they don’t know how to position effectively against a competitor.

In its published form, the workflow is a long doc. To run it for a new competitor, you copy each prompt into a chat window in order and fill in the variables by hand, then copy the output between steps and stitch the final battlecard together yourself. About 40 minutes of orchestration on top of the 20 minutes of strategic thinking that the workflow is supposed to produce.

The same workflow as a /build-battlecard Skill takes about 4 minutes, and the Skill has grown beyond what the published version does. I type two arguments at runtime, the client name and the competitor name, and the Skill does the rest. It reads the client’s CLAUDE.md and messaging files, so positioning and differentiators come pre-filled. From there, it pulls fresh competitor intel from a /competitive-site-monitor Skill that runs on a schedule, plus win rate and pipeline-at-risk numbers from a /run-win-loss-analysis Skill that ran the last quarter’s deals. None of those inputs existed in the original workflow because it asked you to paste them in by hand.

When the Skill finishes, it writes two output files (a 2-page AE quick reference and a detailed deal-strategy version) and pushes both into Notion under the Competitive Center. Then it updates a tracker database (or spreadsheet) so the competitive coverage status stays honest. The published workflow produces one doc. The Skill produces a competitive system.

Converting a workflow into a Skill doesn’t require you to learn SKILL.md syntax. Open Claude and paste it into your workflow, then ask it to convert it into a Skill you can call with a slash command. Or sketch what you want the Skill to do, then paste the existing workflow as source material. Either direction lands you in roughly the same place. Save the file that Claude generates into .claude/skills/, and you’re done. If you’re migrating from ChatGPT Custom GPTs, the same path works. Paste the GPT’s instructions instead of a workflow doc and ask Claude the same question.

The next time you run the Skill, you can tweak it on the fly. Spot a phrase that you don’t like in the output? Tell Claude in the same window. Want a different output format for this client? Mention it. The Skill picks up the correction, and the next run inherits it. A workflow in a doc can’t do that, because every edit is a manual re-save.

I’ll walk through the line-by-line conversion in a future post. The pattern repeats across every Skill I’ve built. My /build-content-calendar Skill grew the same way. It started as a workflow doc. Now it reads the canonical Google Sheet through an MCP and writes the next week’s content row without me touching the chat window.

Pick ten. Don’t write any new ones this month.

If you’ve got a bunch of workflows you’ve written over the past year or two, you don’t need to convert all of them. Pick ten and let the rest stay in the inventory until you need them. Four criteria help you decide which ten go first.

1. The one you’ve run more than five times: Reuse signal beats novelty signal. The workflow that you keep coming back to is the workflow that pays off when it compounds.

2. The one that takes the most copy-paste orchestration to run: The more steps between typing the prompt and getting the output, the more the conversion saves you per run.

3. The one your team or your clients also run: Shared infrastructure beats individual heroics. A Skill that runs the same way for three people is worth converting three times faster than a Skill that only one person uses.

4. The one that touches a system you’d rather not export data from: The workflow that ends with “now go pull the latest deal notes from Gong” or “paste in the last week’s web traffic from GA4” is the workflow with the highest MCP payoff.

“Ten Skills running across three clients is thirty workflow runs a week. The library becomes the process moat.”

— David Sweenor, Founder/CEO, TinyTechGuides

The ten I convert this quarter become infrastructure. The other workflows wait in the inventory, fine where they are, ready for the next time I need one. If you’re stuck on where to start, pick the workflow tied to the angle that your readers can’t get anywhere else. The ten you pick are the ones that turn your library into a process moat.^[6]

The library you already have

Your library of prompt workflows is one conversion away from compounding. The next workflow you write can wait. Pick ten of the ones that you’ve already published and start there.

The full inventory of more than 100 prompt workflows lives at insights.tinytechguides.com. Pick ten this month and convert them. Don’t want to be writing new prompts six months from now? Subscribe, and the next post lands in your inbox.

Share

Leave a comment

Frequently asked questions

What’s the difference between a prompt workflow and a Claude Skill?

A prompt workflow is a linear sequence of prompts that you paste into a chat window, fill in variables for, and run step by step. A Claude Skill is the same workflow packaged into a folder that Claude can invoke with a single slash command. Skills inherit context from the project’s CLAUDE.md and your memory folder, pull data through MCP servers, and pick up corrections from one run to the next. A prompt workflow is one-shot; a Skill compounds.

What is a Claude Skill?

A Claude Skill is a folder of instructions, scripts, and resources that Claude loads when you invoke it with a slash command. Per Anthropic’s documentation, Skills are organized folders that load only when relevant. A Skill typically contains a SKILL.md file describing what the Skill does, optional templates or examples, and pointers to the MCP servers it needs. When you call a Skill in a project, Claude reads the project’s CLAUDE.md and memory folder to fill in context before running the workflow.

How do I convert a prompt workflow into a Skill?

Open Claude, paste your existing workflow, and ask Claude to turn it into a Skill you can call with a slash command. Or start with a sketch of what you want the Skill to do, then paste the workflow in as source material. Either direction lands you in roughly the same place. Save the file that Claude generates into your project’s .claude/skills/ folder. You don’t need to learn SKILL.md syntax by heart; Claude builds the file structure for you.

Which prompt workflows should I convert to Skills first?

Four criteria help you pick. First, the one you’ve run more than five times, because the reuse signal beats the novelty signal. Second, the one that takes the most copy-paste orchestration to run, because the bigger the orchestration, the more the conversion saves per run. Third, the one your team or your clients also run, because shared infrastructure beats individual heroics. Fourth, the one that touches a system you’d rather not export data from manually, because that’s where the MCP payoff is highest.

How are Claude Skills different from Custom GPTs?

Custom GPTs improved on prompt docs by letting you trigger a workflow without needing to find the source file. Skills go further. A Skill reads your project’s CLAUDE.md and memory folder every time it runs, so variables come pre-filled per project. A Skill can call the MCP servers to pull data from Gmail, Sheets, or a CRM. When you spot something you want to change mid-run, you tell Claude in the same window, and the Skill picks up the correction. Updating a Custom GPT requires logging back into the GPT builder.

Why is a library of Skills a process moat?

A library of prompt workflows is dormant inventory. Each entry is one trigger away from working, but nothing compounds across runs. A library of Skills is a process infrastructure. Each Skill inherits from CLAUDE.md and memory, calls tools via MCPs, and carries over corrections into the next run. Ten Skills running across three clients produce thirty workflow runs a week without writing new prompts. The library becomes Hamilton Helmer’s Process Power applied to a marketing function.

About David Sweenor

David Sweenor is a Top 25 AI thought leader, author, and founder of TinyTechGuides. He spent the first half of his career as a practitioner at IBM, working in data science, business intelligence, and data warehousing. In the second half, he led product marketing teams at SAS, Dell Software, Quest, TIBCO, Alteryx, and Alation, covering advanced analytics, AI, and B2B marketing transformation. He writes about AI for marketers, Claude Skills, prompt workflows, and B2B operator depth at TinyTechGuides.

Books

- Artificial Intelligence: An Executive Guide

- Generative AI Business Applications

- The Generative AI Practitioner’s Guide

- The CIO’s Guide to Adopting Generative AI

- Modern B2B Marketing

- The PMM’s Prompt Playbook

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Sweenor, David. “The Marketer’s Case for Claude Code.” TinyTechGuides, May 8, 2026. https://tinytechguides.com/blog/the-marketers-case-for-claude-code/

^[2]Sweenor, David. “The Four Layers of a Claude Stack (and why Skills alone fall apart).” TinyTechGuides, May 13, 2026. https://tinytechguides.com/blog/four-components-claude-stack/

^[3]Sweenor, David. “Marketing Mad Libs: Prompt Variables for Smarter Content Automation.” TinyTechGuides, February 17, 2025. https://tinytechguides.com/blog/marketing-mad-libs-prompt-variables-for-smarter-content-automation/

^[4]Anthropic. “Extend Claude with skills.” Claude Code Documentation, accessed May 15, 2026. https://docs.anthropic.com/en/docs/claude-code/skills.

^[5]Sweenor, David. “Strategic Battlecard Workflow for Competitive Wins.” TinyTechGuides Insights, July 14, 2025. https://insights.tinytechguides.com/p/strategic-battlecard-workflow-for

^[6]Sweenor, David. “The Marketing Moat Nobody Is Talking About in 2026.” TinyTechGuides, May 7, 2026. https://tinytechguides.com/blog/marketing-moat-2026/

The data, analytics, and AI industry is currently obsessed with production velocity. Every vendor is promising that their AI agents can automate workflows, draft emails, order your groceries, and analyze your pipeline in seconds. It sounds great on paper, but when push comes to shove, there’s certainly room for improvement. In my work with clients who are building custom agents, they have some serious concerns and reservations about agentic AI, which are certainly justified. While the agents are fast and dutifully execute tasks assigned to them, they are often confidently wrong more often than not. This occurs because your AI likely has a data-context problem, and context serves as the anchor for accuracy, which agents often lack. This disconnect is reflected in recent research from MIT (2025), which found that 95% of enterprise AI projects fail to deliver measurable P&L impact, often due to a failure to integrate the model with actual business context and workflows.^[1]

When a human being looks at a flawed quarterly business review (QBR) report, they can often spot errors immediately. They understand the business and know that a merger happened last quarter, and that the currency conversion for the EMEA region is manual, and that the “Total Revenue” field excludes services. They understand the relationships between the data and the business outcomes.

AI agents lack this baseline intuition. Without a rich layer of metadata to provide this context, an agent operates as a fast guesser. Sometimes it’s no better than the predictive text capability on my iPhone, which I must admit, is not that great. I recently sat down with Steve Wooledge, CMO at Collate, on the Data Faces Podcast. Steve has spent 20+ years in the datasphere, from Teradata and SAP to leadership roles at Alteryx and Alation. He has seen the hype cycles move from big data to generative AI, and he believes we have reached a shift in how we manage data. To move from experimental AI to reliable, agentic operations, we must treat metadata as the foundational instruction manual for machine intelligence.

About Steve Wooledge

Steve Wooledge is the Chief Marketing Officer at Collate, the company behind the OpenMetadata project. His career spans over 25 years in enterprise sales and marketing leadership at industry giants, including Teradata, SAP, and Business Objects. Steve is a recognized expert in technical product marketing and category creation, having previously built global partner programs at Alteryx and led product marketing at Alation. Outside of the data industry, Steve is a dedicated guitar player with a passion for melodic hard rock and blues.

In our conversation on the Data Faces Podcast, we discuss several hot topics for the agentic era. These include the transition from chemical engineering to product marketing leadership and how to build a “Switzerland” strategy for metadata across multi-vendor ecosystems. We also explore the shift from Data Intelligence to Semantic Intelligence for AI agents and the “Taste Squared” formula for maintaining marketing quality in an automated world.

The evolution of metadata. From inventory to foundation

Metadata spent twenty years as the ignored part of the primordial data stack and remained the least interesting part of the infrastructure. It served as a technical inventory, used to confirm that a specific column was an integer or that a timestamp used a specific format. It served as a technical necessity for database administrators but rarely provided direct, observable business value.

In our conversation, Steve identified three distinct stages in the move toward semantic intelligence. The “Technical Inventory” stage used metadata as a governance checkbox. This evolved into “Data Intelligence,” which is a term popularized by Stewart Bond and companies like Alation that expanded the definition to include the “who, what, where, when, and why” of data.^[2] This stage moved beyond the technical schema to include the operational context of how people used the information.

We are now entering the “Agentic” stage. In this era, metadata is a tool for machines as much as for people. Steve explained that while metadata describes your data, for AI to be accurate and intelligent, it needs that foundational context to prevent hallucinations. If you want an AI agent to pull a report or automate a task, the system must understand the rules and relationships that govern that data. This foundation transforms an automated guesser into an intelligent system.

Semantics. giving AI a “gut feel”

The separation of data from meaning creates the core challenge in modern AI architectures. A human analyst looking at a database sees a column named rev_adj and intuitively understands it refers to a manual revenue adjustment. AI agents require an explicit map to reach that same conclusion. Steve describes this map as “Semantic Intelligence,” which is a framework that provides the digital equivalent of a “gut feel.”

“Semantics is the overall structure and meaning. It includes the relationships between different data elements so an agent can traverse the graph to get at the reason and meaning.”

— Steve Wooledge, CMO, Collate

Technical metadata describes the storage, including the columns, types, and primary keys. Semantic metadata tells you about the intent, such as business rules, KPI definitions, and ontologies. When these relationships are mapped in a graph, an agent can traverse those connections to reason through a query. It can understand that a “Customer” in the CRM is the same entity as a “Subscriber” in the billing system, even if the underlying schemas look different.

By traversing this semantic graph, an AI agent can self-correct. It can recognize when a requested calculation violates a business rule or when a data point lacks the necessary context to be included in a final report. This architectural clarity allows an automated system to operate reliably in a complex enterprise environment. Without this layer, AI projects often get stuck because the models lack the fundamental ability to reason through data dependencies.

The Switzerland strategy. Why AI needs a neutral layer

Enterprise data is inherently messy. Even the most disciplined organizations suffer from fragmented ecosystems, with critical information scattered across Snowflake, Databricks, legacy on-premises databases, and SaaS applications. Each of these platforms offers its own proprietary version of metadata management, and this creates a series of disconnected silos. If your AI strategy relies on the metadata layer of a single platform, you are building on a foundation that cannot see the full picture.

This fragmentation necessitates what Steve calls a “Neutral Layer.” He argues that AI agents require a “Switzerland” strategy: an agnostic metadata layer that sits between the various data silos and the AI models. This neutral layer provides a consistent view of the business logic regardless of where the data lives. It ensures that when an agent asks for “last month’s churn rate,” the definition remains identical whether the data is pulled from a cloud warehouse or a regional database.

“There is no neutral layer that sits across all of that. You need to have this agnostic layer of metadata and semantic intelligence that sits between the data and the AI to ensure you understand the meaning of the information.”

— Steve Wooledge, CMO, Collate

Adopting an agnostic approach also provides a hedge against future architectural changes. As companies undergo mergers, acquisitions, or switch vendors, a proprietary metadata strategy elevates risk and becomes a liability. By using open standards like OpenMetadata, organizations can preserve their semantic intelligence as their underlying infrastructure evolves. Steve’s view is that this neutral layer is the primary way to ensure that your business rules remain portable and your AI remains accurate as you scale across multiple platforms.

Marketing the abstract. Lessons from a first-principles CMO

Marketing a technical product requires a unique level of architectural clarity. If you cannot map the relationships between your data elements, you will struggle to map your message to the specific problems your customers face. Steve credits much of his approach to his time at Business Objects, working under Dave Kellogg, who is a veteran leader who preached the power of first principles. This philosophy dictates that marketing exists to reduce friction in the sales process by grounding every message in clarity and logical sequence.

When you sell an abstract concept like a “metadata platform,” you cannot lead with features. A CFO or CEO rarely wakes up thinking about their cataloging needs. Instead, you must sell the business outcomes that metadata enables, such as AI safety, operational velocity, and what we call consequence management. By visualizing the invisible through semantic metadata graphs, marketers can make these complex technical structures tangible for executive budget owners.

This first-principles approach also fuels grassroots expansion through open-source communities. By allowing developers to solve immediate technical problems using tools like OpenMetadata, a company can build a foundation of trust before moving toward an enterprise-wide engagement. Steve’s experience at Alation and Alteryx confirms that when you give people the tools to prove value in their own environment, the transition to a strategic partnership becomes a logical next step.

The “taste squared” era of marketing

Velocity without judgment is just noise. The integration of AI into marketing workflows has fundamentally changed the expectations for production velocity. We can now develop content, campaigns, and landing pages at a pace that was previously impossible. However, this increased speed introduces a risk that Steve calls “lazy marketing.” While AI can generate high volumes of content, it often lacks the subtlety and judgment required to connect with a specific customer base.

To address this challenge, Steve references a formula popularized by Tom Wentworth^[3], where marketing output equals AI adoption multiplied by taste squared. This perspective suggests that while adopting AI is a linear requirement for modern teams, human taste acts as an exponential multiplier for quality. Having the technical skill to use a prompt is one thing, and having the taste to know when a message is great, and when it is “average AI,” is what will differentiate the leaders from the laggards.

Maintaining this level of quality requires a commitment to the human element of marketing. In our conversation, Steve emphasized that you still have to “slave over the word” to ensure your message correctly lands. This means using AI as a tool for acceleration rather than a replacement for thinking. By combining automated velocity with rigorous peer review and high creative standards, marketing leaders can use AI to amplify their impact without sacrificing brand integrity.

The infrastructure of trust

Building an AI strategy without a solid metadata foundation is like attempting to build a penthouse on a swamp. The agents you deploy will only be as intelligent as the context you provide them. By adopting a neutral, semantic metadata layer, organizations can equip their AI systems with the digital intuition needed to move beyond simple automation and toward autonomous operations.

Metadata is the primary architectural anchor of the agentic era, supporting both governance and agentic AI. To learn more about how to build this foundation for your own organization, you can listen to the full conversation with Steve Wooledge on the Data Faces Podcast and explore the open-source community at OpenMetadata.

Listen to the full conversation with Steve Wooledge on the Data Faces Podcast.

Based on insights from Steve Wooledge, CMO at Collate, featured on the Data Faces Podcast.

Share

Leave a comment

Frequently asked questions

What is the difference between metadata and semantics? Metadata describes the technical properties of data, such as column names, data types, and timestamps. It acts as a technical inventory of information. Semantics represents the overall structure, meaning, and relationships between those data elements. While metadata tells an AI agent what a field is, semantic intelligence tells the agent how that field relates to business rules and KPIs across the enterprise.

Why do AI agents need a neutral metadata layer? Most organizations store data across fragmented ecosystems like Snowflake and Databricks. Each platform manages metadata in its own proprietary way. A neutral metadata layer sits between these silos and the AI models, providing a consistent, agnostic view of business logic. This strategy ensures that an agent’s understanding of the data remains accurate even if the underlying infrastructure changes.

How does semantic intelligence prevent AI hallucinations? AI hallucinations often occur because the model lacks the necessary context to interpret data correctly. Semantic intelligence provides a mapped graph of relationships that allows an AI agent to reason through a query like a human analyst. By traversing this graph, the agent can identify when a requested calculation violates a business rule or when it lacks the context required for an accurate response.

What is the taste squared formula for marketing? CMO Tom Wentworth introduced the formula. Marketing output equals AI adoption multiplied by taste squared. It suggests that while AI adoption is a linear requirement for productivity, human taste is an exponential multiplier for quality. In an era where anyone can use AI to generate average content, the differentiator for marketing leaders is the judgment required to refine AI output into something resonant.

Podcast highlights

• [0:00] Introduction to Steve Wooledge and Collate

• [1:08] The journey from chemical engineering to technical data sales

• [3:45] Melodic hard rock and guitar shredding as a creative outlet

• [5:01] Lessons from Dave Kellogg on first-principles marketing

• [7:40] The reality of partner marketing with global system integrators

• [10:18] Why open-source projects out-innovates proprietary enterprise software

• [15:56] The shift from technical metadata to semantic intelligence for AI agents

• [20:45] Building a Switzerland approach to metadata across multi-vendor silos

• [24:03] How AI velocity is fundamentally changing the CMO role

• [27:23] The taste squared formula and why you cannot be a lazy marketer

• [32:19] Career advice for the next generation of data and marketing professionals

• [36:28] Final advice on peer review and maintain quality control

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]MIT Project NANDA. 2025. “The GenAI Divide: State of AI in Business 2025.” MIT Sloan Management Review.

^[2]Bond, Stewart. 2026. “Your AI has a data intelligence problem.” TinyTechGuides.

^[3]Wentworth, Tom. 2024. “AI Adoption and the Taste Square.” incident.io.

When I was writing the last post on the four components of a Claude stack, I went to grab a screenshot of my memory folder and couldn’t find it.^[1] Finder didn’t show .claude and Spotlight returned nothing. For ten minutes, I was frantically trying to figure it out. Did I have Mad Cow? Amnesia? Did Claude and I delete the whole thing when I mindlessly clicked continue when it prompted me to do so, as it was writing some Python code? Fortunately, not, Mac was just hiding it from me.

Anyone who read that post and went looking for the folder may have encountered the same issue. The folder exists, but it’s not in the repo you’re working from or where Finder wants you to look. The Memory component is one of the four pieces that make a Claude marketing OS work, and it lives in a part of your home directory that Mac considers off-limits by default. Finding the folder matters because it’s where everything that compounds across your sessions lives. Claude writes to it, you curate it occasionally, and over time, it holds more of your operating knowledge than any other file on disk.

An earlier post, “The marketer’s case for Claude Code,” covered the install and the four pieces.^[2] The follow-up showed how those pieces compound. This post answers where the Memory component lives and how to open it without the ten minutes of panic that I just spent.

Here’s a five-minute walkthrough to help Mac users overcome some of the “smart defaults” that were put in place to simplify things but end up causing agita. Windows is easier, because File Explorer doesn’t auto-hide files by default. Two simple steps will cure all ills.

Where Claude stores your memory on a Mac

Every project gets its own memory folder inside your home directory. The base path on a Mac is ~/.claude/projects/<encoded-project-path>/memory/. Anthropic provides an easy way to open it without typing the path yourself. From inside Claude Code, type /memory and you’ll get a list of the memory files loaded into your session, along with a link to the folder.^[3] Make sure you select the “Open auto-memory” folder option, which opens the folder in Finder.

But the /memory command doesn’t put the folder in the Cursor’s file tree, which is where I work. Each time I want to read a saved file, running a slash command is one too many steps. There’s a better answer, but first you need to know why Mac hides this folder from you.

Mac hides any file or folder whose name starts with a dot. By default, Finder ignores them, so your user folder doesn’t look like a server room. The Claude folder is called .claude, and Spotlight ignores the same files. That’s why my search for “claude” returned nothing during my ten-minute panic.

Finder will reveal hidden files if you tell it to. Open your home folder, then press Cmd+Shift+. (Command, Shift, and the period key together). Hidden files and folders appear. Press it again to hide them. Inside .claude, you’ll find a projects/ folder, and inside that, a subfolder named after your project’s full filesystem path, with slashes replaced by hyphens. Clunky, but you only need to navigate it once before the next move fixes it for good.

A symlink does the heavy lifting. One terminal command places a shortcut to the memory folder inside your project, so the link shows up at the top of your project’s file tree in Cursor. The full command looks like this:

ln -s ~/.claude/projects/-Users-david-sweenor-Documents-TTG-Advisor/memory ~/Documents/TTG-Advisor/.claude-memory

Replace the paths with your own project name and folder location. Or, if you’d rather not type the command yourself, ask Claude to create the symlink for you! Claude will run it with the correct paths.

A quick aside for the non-terminal natives. If you didn’t start your career at IBM running vi and Emacs on UNIX boxes, you may not have seen this one before. A symbolic link, or symlink, is a UNIX-era trick. The command creates a file that acts like the folder that it points to. Click on the link in Cursor, and you’re inside the memory folder, even though the actual files live in your home directory.

The link is named .claude-memory with a leading dot, so Finder still hides it. Cursor and most editors show dotfiles by default, so the link is visible exactly where you need it. If you’d rather see it in Finder too, drop the dot. Name it claude-memory instead.

Windows readers, the path resolves identically. C:\Users\<you>\.claude\projects\... puts you in the same folder structure. File Explorer doesn’t hide dot-files on Windows, so .claude is visible the moment you open your user folder. Different OS philosophy, same outcome. You can already see it.

What you’ll find inside

Once you’ve found the folder, you have to decide what’s worth putting in it. My TTG memory folder currently holds more than 60 files. Every entry is a decision about what survives across sessions.

The naming convention is a topical prefix followed by a short descriptor: feedback- for working-style rules and corrections, project- for active client work, reference- for pointers to external systems, and topic-specific prefixes like podcast-, wordpress-, and youtube- for tooling notes that don’t fit elsewhere.

Three real entries from my folder:

- feedback-flagged-ai-phrases.md: phrases I keep catching in drafts that need to die (”game-changer,” “shift” as a noun, and “delve”)

- reference-email-signature.md: the exact block to put at the bottom of every outbound email

- <obscured-client-project>.md: what’s active with one of my clients right now, who owns what, and what’s due next

MEMORY.md sits at the root of the folder. It’s the index, and each topic file gets a one-line entry. Claude reads the first 200 lines (or 25KB, whichever comes first) at the start of every session.^[4] The index loads into every conversation; the topic files load only when Claude needs them. Think of MEMORY.md as the table of contents and the topic files as the chapters.

Curation cuts both ways. I don’t save session-specific task state, for instance. I tried it once, and a note about which post I was drafting on a Tuesday in March was useless by Wednesday morning. Now I save what still matters next week.

Memory is the piece that survives

Curation matters only because what you save compounds. Of the four components from the last post, Memory has the longest half-life. Skills run the plays that you trigger. CLAUDE.md captures each project’s rules. MCPs connect Claude to your live tools. Memory does what the others can’t. It carries what you’ve learned across every session, every project, and every conversation that follows.

Claude saves your corrections once. You don’t have to repeat them next session. The preferences that you record stay recorded. Over time, the folder shapes how Claude responds to the specific work that you do. As I’ve written before, teaching AI in your context beats generic automation.^[5]

What’s worth saving and what isn’t is its own conversation. I’ll cover it in a future post on my weekly Claude playbook for B2B marketers. For now, settle for finding the folder and placing a symlink so Finder never traps you again.

Ten minutes of mad-cow panic later, I had my folder back. The harder question is whether I’ll use it once I know where it lives.

Share

Leave a comment

Frequently asked questions

Where does Claude Code store memory files on a Mac?

Claude Code stores memory files at ~/.claude/projects/<encoded-project-path>/memory/. Each project gets its own subdirectory inside .claude/projects/, named after the project’s filesystem path with slashes converted to hyphens. The .claude folder lives in your home directory but is hidden by default because it starts with a dot. The base path resolves to C:\Users\<you>\.claude\projects\... on Windows, where File Explorer shows it without any reveal step. Linux uses the same ~/.claude/projects/... path as Mac, with the same dot-file hiding convention.

Why can’t I see the .claude folder on my Mac?

macOS hides any file or folder whose name starts with a dot. Finder ignores them by default, and Spotlight skips them too. To reveal hidden files, open your home folder in Finder and press Cmd+Shift+. (Command, Shift, and the period key). The .claude folder appears in the file list. Press the same shortcut again to hide hidden files. Windows File Explorer doesn’t apply this convention, so .claude is visible there without any reveal step.

How do I make the Claude memory folder visible inside my project?

A symbolic link works best. From your terminal, run ln -s ~/.claude/projects/<your-encoded-path>/memory ~/Documents/<your-project>/.claude-memory. The symlink appears at the top of your project’s file tree in Cursor and other editors that show dotfiles by default. You can also ask Claude to create the symlink with the right paths substituted. The folder name keeps a leading dot, so Finder still hides it; drop the dot if you want to see it there too.

What is the /memory slash command in Claude Code?

The /memory command shows a list of memory files loaded into your current Claude Code session and provides a link to open the auto memory folder. Run it from within Claude Code, then click the link to launch the folder in Finder. The command works on every operating system that Claude Code supports, including Mac, Windows, and Linux. It’s the official Anthropic way to find the folder without typing the path, but it doesn’t place the folder inside your project’s file tree where you work.

What is MEMORY.md and what gets loaded into a Claude session?

MEMORY.md sits at the root of the auto memory folder and acts as the index. Each topic file in the folder gets a one-line entry in MEMORY.md. At the start of every Claude Code session, Claude reads the first 200 lines (or 25KB, whichever comes first) of MEMORY.md. The topic files themselves load only when Claude needs them. Think of MEMORY.md as the table of contents and the topic files as the chapters Claude pulls when relevant.

What’s the easiest way to open Claude Code’s memory folder?

Three paths work. Run /memory from inside Claude Code and click the link to launch the folder in Finder. That’s the fastest one-off method. To browse manually, open Finder, press Cmd+Shift+. to reveal hidden files, and navigate to ~/.claude/projects/<your-project>/memory/. For daily access while editing, create a symbolic link within your project that points to the folder so it appears in the Cursor’s file tree. The /memory command works on every operating system; the symlink is most useful when you live inside an editor like Cursor.

About David Sweenor

David Sweenor is a Top 25 AI thought leader, author, and founder of TinyTechGuides. He spent the first half of his career as a data practitioner at IBM, working in data science, business intelligence, and data warehousing, and the second half in product marketing leadership at SAS, Dell, Quest, TIBCO, Alteryx, and Alation. His writing focuses on the practical intersection of AI, analytics, and B2B marketing.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications

- The Generative AI Practitioner’s Guide

- The CIO’s Guide to Adopting Generative AI

- Modern B2B Marketing

- The PMM’s Prompt Playbook

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Footnotes

^[1]Sweenor, David. “Is your Claude marketing OS a little quirky?” TinyTechGuides, May 13, 2026. https://tinytechguides.com/blog/four-components-claude-stack/

^[2]Sweenor, David. “The marketer’s case for Claude Code.” TinyTechGuides, May 8, 2026. https://tinytechguides.com/blog/the-marketers-case-for-claude-code/

^[3]Anthropic. “How Claude remembers your project.” Claude Code documentation. Accessed May 13, 2026. https://code.claude.com/docs/en/memory

^[4]Anthropic. “How Claude remembers your project.” Claude Code documentation. Accessed May 13, 2026. https://code.claude.com/docs/en/memory

^[5]Sweenor, David. “The Barcode on the Bronze: Why Your AI Needs to Know What Makes You Different.” TinyTechGuides, November 18, 2025. https://tinytechguides.com/blog/the-barcode-on-the-bronze-why-your-ai-needs-to-know-what-makes-you-different/

Many of my clients are gravitating towards Claude Code for marketing. This is surely a step in the right direction. However, most of them are not widely using the Skills that you can create with Claude. In fact, I’ve only recently understood their power and how to use them to their fullest extent. I’ve built a /podcast-prep skill, a /podcast-production skill, a /send-invoice skill, along with many others. In total, I’ve created 60+ of them across two working folders, one for my TinyTechGuides business and another for client work. When I first started using skills, they mostly worked, but at times, they sometimes didn’t consistently do what I wanted. Little did I know their true power.

One of my skills is /build-content-calendar, which is pretty effective. Unfortunately, after using it extensively, I started noticing something off. I’d kept the content calendar in two places, a local CSV and a Google Sheet. Some skills updated the CSV, while others updated the Sheet. When I noticed the discrepancies, I would ask Claude to sync or fix them by hand and move on. A few days later, the two files were out of sync again. The skills had run fine both times. They each edited whichever file they knew about. Turns out, I broke a cardinal rule: one source of truth — I should have had the local copy simply reference the Google Sheet version.

“I fixed it. A few days later, it was broken again.”

The same week, I caught my /podcast-prep skill churning out question lists with em-dashes I’d banned, and it kept opening with generic “tell us about yourself” questions even though I’d told it to always pull from the guest’s recent LinkedIn posts. That’s when I realized the skills were doing exactly what they were told to do. Nothing more and nothing less. What was missing? Well, I only had part of the recipe — nothing was telling the Skills what was true here. Three more components fixed it.

A Skill on its own is one of those prompt workflows I worked on last year.^[1] What’s the secret? Well, if you combine four components, most of the quirks you keep noticing stop being structural in nature. Here’s what an effective Claude stack looks like, and what each component does:

• Skills run the plays: packaged workflows you trigger with a slash command, like /podcast-prep, /send-invoice, or /podcast-production.

• CLAUDE.md sets the rules: the project’s rulebook, which Claude reads at the start of every session in that folder.

• Memory holds the truth: the personal preferences, corrections, and decisions that follow you across every project.

• MCPs connect the tools: live connectors into Gmail, Sheets, Calendar, WordPress, and the rest of your stack.

Each ingredient on its own is just an ingredient. Combined in the right kitchen, they compound into a working stack.

A Skill alone is a prompt workflow

Earlier this month, I made the marketer’s case for Claude Code.^[2] That post explained why marketers should run Claude as a project tool. This piece covers the four components that make it stick once you do.

A Skill is a packaged recipe Claude calls with a slash command. It’s an atomized set of instructions for whatever you want to do in marketing. The recipe lives in a folder, alongside any helper scripts or examples it needs.^[3] You type /podcast-prep, Claude loads the instructions in the folder, and the skill runs. Type /send-invoice and you get a different recipe. The skill is the “how” of a repeated job. The great thing about them is that they load when you call them, not in every session. Why do you care? It saves tokens and keeps the context window from getting too bloated.

In isolation, that’s all it is. A Skill is a recipe with no kitchen, no pantry, and no cook’s preferences. The recipe says, “build a guest prep doc in this format.” And as we know, every recipe has the ingredients and the steps you need to follow to create that delicious meal. It does not say which guest you’re prepping for, whether the questions should lean technical or strategic, or which generic openers I’ve already banned. If you run the same /podcast-prep skill for two different guests, you’ll get the same mistakes both times. Nothing provides context or tells the skill what’s true about this guest, this episode, or what I want to avoid.

That’s where CLAUDE.md comes in. CLAUDE.md is the project’s rulebook, and Claude reads it every time you start a session in that project’s folder. The Skill brings the workflow, and the CLAUDE.md tells the Skill what’s true in this room. Together, they produce a coworker. Without CLAUDE.md, you’ve got a contract writer who hasn’t read the brief, which contains the bans, link conventions, and project-specific rules.

In this project, my CLAUDE.md tells the /podcast-prep skill to pull guest questions from the guest’s recent LinkedIn posts and prior podcast appearances, not generic conference talking points. It tells the skill that em-dashes are banned in the prep doc, that guest titles must match what’s on LinkedIn (not what’s on the company website), and that every question list opens with something tied to the guest’s recent work. The same /podcast-prep skill ran for two different guests, producing two different sets of prep notes: same recipe, different room.

“A Skill without CLAUDE.md doesn’t know which client it’s working for.”

Which fixes one half of the brittleness. CLAUDE.md handles the project. The other half is what travels with you across projects.

A CLAUDE.md alone is a wiki nobody reads

CLAUDE.md is the project-scoped rulebook. Drop a file named CLAUDE.md at the root of a project, and Claude reads it every session you start in that folder. The file contains voice rules, naming conventions, and project-specific bans such as “no em-dashes” or “no Amazon book URLs.”

It’s also useless on its own. A 500-line rulebook is wallpaper if nothing reads it, and rules in a file you can’t trigger don’t bind anything. Without memory, the rules stop at the project’s edge. CLAUDE.md says “first person, sentence case headings.” It does not say I prefer “leadership” over “the board.” That preference is mine, and it follows me across every client folder I open.

“Without memory, every CLAUDE.md is the same lecture I have to repeat in every project.”

Memory does the inverse of CLAUDE.md. Memory is what Claude remembers about how you actually work in this project, across every session. It holds your voice preferences, the decisions you’ve made about how to phrase things, and the file naming conventions you’ve earned the hard way. Each project gets its own memory folder, which lives outside the repo in your home directory, and Claude reads it at the start of every session.

In this project, the CLAUDE.md handles the format rules. Memory adds the word-level preferences I’ve corrected too many times to trust myself to remember. Project rules and personal preferences live in different files for a reason. Together, they steer the voice every time I work in this folder.

Some teams write the same file as AGENTS.md instead.^[4] It’s the cross-tool convention stewarded by the Linux Foundation’s Agentic AI Foundation, read natively by Codex, Cursor, and 20-plus other tools. Anthropic’s tooling reads CLAUDE.md, and some projects keep both files.

Project rules are sorted. Personal preferences live in a different file. That file has its own way of disappointing you when it works on its own.

Memory alone is sticky notes taped to your monitor

Memory in Claude is a folder of markdown files that the system reads at the start of every conversation, regardless of project. Mine has more than fifty entries. Here are five of them, lightly anonymized:

Sheet 1BojY1g is canonical for the calendar. Don’t edit the local CSV.

I prefer “leadership” over “the board.” Applies in every project.

Pull quotes attribute to “David Sweenor, Founder/CEO, TinyTechGuides.”

Buffer’s createPost API does not ingest video from URLs. Manual upload only.

Don’t update episodes.md or guest-list.csv until after the episode is recorded. Guests cancel.

Five real entries. They’re project-agnostic. They follow me from this TTG repo into a Posit project into a Solidatus project, every time I open Claude. Memory is what Claude knows about me, regardless of what I’m working on. Of course, I also have a memory for each of my client folders and, in some cases, client projects.

It’s also sticky notes on the monitor. The notes are useful when I’m sitting at the desk reading them. They don’t open Notion or send the Buffer post. They certainly don’t check what’s next on the Sheet. Memory tells Claude what I’ve decided. Acting on those decisions still takes me reaching for the keyboard. It’s the kind of bugaboo that looks fixed every time you stare at it.

Memory holds the Sheet ID, but something else has to actually read the Sheet. Memory remembers I want video posts queued in Buffer, but something else has to post them. The memory entry is just text, Claude can’t act on it without a connector to the tool itself.

“Memory tells Claude what you decided. MCPs let Claude do something about it.”

Which inverts the question. What does an MCP look like as the only component? It is loud, connected, and not very useful.

An MCP alone is an API call you’re making by hand anyway

Last quarter, I asked Claude to track down a follow-up email I’d sent a prospect two weeks earlier. The Gmail MCP happily searched 200 messages and returned 47 of them. None were the thread I was chasing.

That’s an MCP doing exactly what an MCP does. An MCP, or Model Context Protocol server, is a connector that lets Claude operate tools like Gmail, Sheets, and Calendar.^[5] Anthropic introduced the spec in late 2024, and it’s now the standard interface for AI applications to talk to external systems. If you’ve ever wanted Claude to send a follow-up email instead of drafting one for you to copy-paste, MCPs are how.

On their own, MCPs are raw access. The Gmail MCP can search your inbox. The Sheets MCP can read any range. Neither one knows which thread is the prospect call you’ve been chasing or which tab is the canonical source of truth. Raw access without instructions is “search 200 emails for X” on repeat, which is mostly the API call you’d be making by hand. You’ve just stuck Claude in the loop.

That’s where Skills come back in. A Skill is a recipe that turns “search 200 emails for X” into “find the threads with my recent prospect contacts and draft three follow-ups in my voice.” The Skill carries the workflow. The MCP carries the tool access. Skills plus MCPs is the difference between an API call and a finished workflow.

“An MCP without a Skill is a kitchen without a chef.”

My /podcast-production skill bundles ten clip drafts and queues them in Buffer in about ninety seconds. Each draft has the correct ATTACH header, so I know which MP4 to upload manually. The skill knows the format from CLAUDE.md, the Buffer account from memory, and the API call from the MCP. Pull any component, and the workflow stops being a workflow.

What the four components buy you

By the time you’ve stacked all four, you’ve built an operating system. Skills hand work to CLAUDE.md, which gives memory the rules. Memory feeds MCPs, which serve the next Skill. The output of one component becomes the input to the next in every session.

Consider the following questions:

1. Are your Skills reading the project rulebook, or are they running on default voice?

2. Does your CLAUDE.md exist in every project you switch between, or just the one you set up first?

3. Have you corrected the same preference more than twice this month? That’s a memory miss.

4. Are you still hand-copying data from Gmail, Sheets, or your CRM into chat?

Gartner put a sharper point on it at the D&A Summit in Orlando this March. Expecting AI to compensate for delayed upgrades, siloed teams, and missing context is “wishful thinking.”^[6] Gartner prescribes “a well-designed context layer” instead. The four components in this article are that context. The Skill is the tool that runs on top. The compounding makes them a marketing moat.^[7] Tools change, and components compound.

The next time your Claude setup feels like it’s not working quite right, look down. Whatever Skill you wrote is doing its job. The job is just bigger than one component.

More on the four-component stack every other Tuesday (or when I get around to writing about it) at tinytechguides.com or insights.tinytechguides.com. Skills and CLAUDE.md, memory and MCPs, and what breaks when any one of them goes missing. Subscribe if you want the next one in your inbox.

Share

Leave a comment

Frequently asked questions

What is a Claude marketing OS?

A Claude marketing OS — sometimes called a Claude stack — is the four interlocking components that make Claude reliably useful for marketing work: Skills (slash-command recipes), CLAUDE.md (project rulebooks), memory (personal preferences across projects), and MCPs (connectors to tools like Gmail and Sheets). When the OS feels “quirky,” one of the four is missing or misaligned. Stacked together, they compound. The output of one component becomes the input for the next.

What is a Claude stack?

A Claude stack is the four interlocking components that make Claude reliably useful for repeated work: Skills (slash-command recipes), CLAUDE.md (project rulebooks), memory (personal preferences across projects), and MCPs (connectors to tools like Gmail and Sheets). Each component alone is limited. Stacked, they compound. Skills hand work to CLAUDE.md, CLAUDE.md gives memory the rules, memory feeds MCPs, and MCPs serve the next Skill. The output of one component becomes the input for the next.

What’s the difference between a Skill and a CLAUDE.md?

A Skill is a packaged recipe Claude calls with a slash command, like /podcast-prep or /send-invoice. The Skill carries the workflow steps. A CLAUDE.md is the project’s rulebook, located at the root of the project folder, and Claude reads it every session in that project. The CLAUDE.md file tells the Skill what’s true in this project, such as voice rules, naming conventions, and project-specific bans. Skills handle the “how.” CLAUDE.md handles the “what’s the context here.” Without CLAUDE.md, the Skill runs on the default voice.

What is an MCP in Claude?

An MCP, or Model Context Protocol server, is a connector that lets Claude operate the tools you already use, like Gmail, Sheets, Calendar, Notion, and Buffer. Anthropic introduced the spec in late 2024, and it’s now the standard interface for AI applications to talk to external systems. On its own, an MCP is raw access, like “search 200 emails for X” without context for which X matters. When paired with a Skill, the MCP becomes a finished workflow.

Why won’t a better Skill alone fix my Claude workflow?

Skills carry instructions for repeated tasks, but they don’t include project context, personal preferences, or access to tools on their own. A Skill running without CLAUDE.md doesn’t know which client it’s working for. Without memory, it forgets your preferences from session to session. The same Skill without MCPs can’t act on your tools at all. The Skill executes whatever you tell it. If nothing else in the stack tells it what’s true, the executions stay shallow. Sharper Skills don’t fix missing components.

Where should I start auditing my Claude stack?

Start with the four-question audit. Are your Skills reading the project rulebook, or are they running on default voice? Does your CLAUDE.md exist in every project you switch between, or just the one you set up first? Have you corrected the same preference more than twice this month? That’s a memory miss. Are you still hand-copying data from Gmail, Sheets, or your CRM into chat? Each “no” maps to a missing component.

Is AGENTS.md the same as CLAUDE.md?

Same idea, different filenames. AGENTS.md is the cross-tool convention stewarded by the Linux Foundation’s Agentic AI Foundation, read natively by Codex, Cursor, Gemini CLI, and 20-plus other tools. CLAUDE.md is Anthropic’s tooling convention. Both serve as project rulebooks that the agent reads at the start of each session. Some teams keep both files for cross-tool portability. The function is a project-scoped context layer, which matters more than the filename.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Sweenor, David. “PMM’s Prompt Playbook - Prompt Inventory.” insights.tinytechguides.com, February 21, 2025. https://insights.tinytechguides.com/p/pmms-prompt-playbook-prompt-inventory

^[2]Sweenor, David. “The marketer’s case for Claude Code.” TinyTechGuides, May 1, 2026. https://tinytechguides.com/blog/the-marketers-case-for-claude-code/

^[3]Anthropic. “Agent Skills Overview.” Claude API Docs. Accessed May 2026. https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview

^[4]Agentic AI Foundation. “AGENTS.md.” Linux Foundation, 2026.

https://agents.md/

^[5]Anthropic. “Introducing the Model Context Protocol.” November 25, 2024. https://www.anthropic.com/news/model-context-protocol

^[6]Gartner. “Gartner Identifies Three Pillars for Deriving Value from AI.” Press release, March 9, 2026. https://www.gartner.com/en/newsroom/press-releases/2026-03-09-gartner-identifies-three-pillars-for-deriving-value-from-ai

^[7]Sweenor, David. “Marketing moats: what of that?” TinyTechGuides, May 8, 2026. https://tinytechguides.com/blog/marketing-moat-2026/

Everyone’s talking about moats

Well, it’s happened. The AI world is suddenly abuzz and enamored with moats, but these are not the stone walls and fetid water that surrounded medieval castles. These are metaphorical moats that provide a systemic competitive advantage. It seems like half of LinkedIn has rediscovered Hamilton Helmer this quarter. Software Equity Group, an M&A advisory firm that has tracked the SaaS sector for thirty years, reports that 85% of M&A buyers now name AI-driven commoditization as the number-one risk to SaaS valuations.^[1] The high-priced consultants got the memo, and brand is now the new moat. No wait, trust is the moat. Or perhaps, data flywheels are the moat? Every 2026-predictions post has a take, and most of those takes have some truths.

Dave Kellogg is right about trust. “Only trust will get people to open your emails,” he wrote in his 2026 predictions, “only trust will allow them to believe the reviews and testimonials about your product.”^[2] CTI Digital is right about brand in a sea of AI-generated sameness.^[3] Steven Cen is right about data flywheels.^[4] The consensus has converged for a reason.

Most of that writing is shaped by capital-market concerns. What investors want to know is what defends a SaaS company’s valuation when they get nervous about the impending AI bubble popping. That is a useful question for boards, founders preparing to sell, and PE shops doing diligence. For CMOs and marketing leaders trying to plan the next quarter, it’s not all that useful.

If you run a marketing function in 2026, “build a data flywheel” and “invest in brand” are not action items for next quarter. They’re often a multi-year investment and require diligence, a concerted effort, and likely, a considerable investment. These are real moats. You can start building them this Tuesday, but sadly, they will not be completed within the next cycle.

A related question that marketing leaders keep asking is: what can a marketing team build in the next quarter that is hard for a competitor to copy? Flywheels, brand, and trust moats in 2026 are outputs. This post is about the input.

Helmer’s forgotten power

Hamilton Helmer published 7 Powers in 2016. It is the strategy book Andreessen Horowitz hands to founders, and the closest thing the post-Porter generation has to a canonical framework.^[5] The book lists seven sources of durable competitive advantage: Scale Economies, Network Effects, Counter-Positioning, Switching Costs, Branding, Cornered Resource, and Process Power.

“I spent eleven years on a fab floor watching Process Power get built one wafer at a time. The machines to do it in marketing landed in 2024.”

— David Sweenor, Founder/CEO, TinyTechGuides

Six of the seven powers belong to the company and the CMO cannot act on them. Scale economies, switching costs, and network effects are not marketing decisions. You do not get to choose your scale curve or your churn dynamics from a marketing seat.

Process Power is the exception. Helmer defines it as “embedded company organization and activity sets which enable lower costs and/or superior product, and which can be matched only by an extended commitment.”^[6] His example is the Toyota Production System (TPS).

In the semiconductor world, we lived by a specific mantra: If the yield drops, the process failed you. We didn’t blame the operator; we fixed the system. Is the line losing wafers to defects? You use data to find the cause, codify the fix into the standard operating procedures, and ensure the next shift doesn’t rediscover the same failure.

In 2026, the marketing function will finally have the instruments to run the same playbook. It’s a mental shift from “campaign thinking” to “platform thinking.”

The operator stack is tool-neutral

A common critique of building an “AI moat” is platform risk. If you build your business on a single model, you’re building a house of cards. But the Operator Stack isn’t about the tool; it’s about the schema of your business logic. Whether you run on Claude, Gemini, or a local open-source model, the moat is the codified practice sitting in your markdown files.

The stack consists of four interlocking layers that behave like a production line:

• Skills: are the named, reusable workflows your team calls by command. A Skill bundles the quality bar, so the output is consistent, regardless of who (or what) is running the prompt.

• Rules (CLAUDE.md): are the standard work documents, distinct from voice guidelines or a brand brief. They capture what gets done every time and what never gets done. When a piece of content misses the mark, you don’t just edit the text. You update the Rules so the failure never repeats. The process failed you, so you fix the process.

• Memory: is the learned-context layer. It’s where your authenticity and unique data live. It turns your napkin files and messy first-party observations into a compounding asset.

• MCPs (Model Context Protocols): are the connective tissue. They remove the “hand-off tax” between the AI and your CRM, content sheet, or inbox. This is where you eliminate friction.

The human as system pilot

The key with all of this is to augment human ingenuity and capacity. You move from being a “Writer” to a System Pilot. Your job is to keep the stack fresh. At TinyTechGuides, I use a /reflect skill to audit the last ten sessions and identify where the rules are drifting or where the memory needs a “napkin file” update from the latest sales call.

A “Writer” is a commodity in 2026. A System Pilot is a high-value strategist who manages a compounding asset (the stack) to produce 10x the output at 10x the quality.

The human provides the “yield” by challenging the AI, conducting research, and feeding the system new, authentic inputs. The stack carries the load, but the pilot determines the destination.

“We do not have a content team. We have a content stack. The same person produces four times the output at the same quality bar, because the stack carries the load—the voice, the data, the authenticity—that used to live in heads.”

— David Sweenor, Founder/CEO, TinyTechGuides

Why this compounds faster than brand

Brand compounds. Trust compounds. Data flywheels compound. The operator-depth camp does not disagree with any of that. The argument is about cycle time.

“Brand compounds in five years. Operator depth compounds much more quickly. The math is not subtle.”

— David Sweenor, Founder/CEO, TinyTechGuides

Brand compounds over a five-year cycle, often with an eight-figure paid spend behind it. Most marketing functions are not catching incumbents on brand. The CMO who needs differentiation in 2026 is not waiting until 2031.

Trust compounds more slowly than brand. A B2B SaaS company earns trust through consistent shipping, customer outcomes that hold up under scrutiny, and reviews from named buyers. That work pays back, but it pays back over a five-to-ten-year arc. That is the arc Kellogg is talking about when he says only trust will get people to open your emails. It is right, and it is slow.

Data flywheels compound, once you have the data. Most teams do not. The flywheel works once you already have customer volume and the feedback loop in place. For an early-stage marketing function, “build a data flywheel” is closer to “build a customer base” than to “execute next week.”

Operator depth compounds on a ninety-day to six-month cycle. The CLAUDE.md you write this weekend starts paying back next Tuesday. The third Skill your team converts from a one-off prompt is sharper than the first, and the fifth is sharper than the third. The memory file accumulates, and the MCPs harden against the systems your team runs every day. By month six, the team is moving at a tempo competitors cannot match without doing the same work, which most of them are not yet doing.

This is why operator depth is the moat available to teams without a ten-year head start. The other moats favor incumbents. Operator depth is the input that makes the consensus moats reachable.

What CMOs should do this quarter

A CMO can start building operator depth in ninety days by running four moves in sequence. None require new headcount, new vendors, or a budget conversation with finance. This is the tactical version of the strategic posture.

“You do not need ten years. You need ninety days. The CLAUDE.md you write this weekend is the standard work document for everything your team ships next quarter.”

— David Sweenor, Founder/CEO, TinyTechGuides

Here are some practical steps you can take today:

Inventory the recurring work: Walk through what your team did last quarter. Anything the team executed more than three times is a Skill candidate. Every product-launch announcement, every analyst-relations note, every webinar promo qualifies. Pick the three highest-frequency workflows and write them as Skills first. The math is mechanical. If a workflow runs ten times a quarter and a Skill saves an hour each run, the Skill pays for itself the second time it runs.

Write a real CLAUDE.md: Not voice guidelines or operating rules. What every piece of content does every time, what no piece of content ever does, and which competitors are off-limits to cite. Treat it as the standard work document for the marketing line. The first version will be wrong in places. By the third pass, it starts saving you time.

Make memory a deliberate practice: Capture editorial preferences, brand decisions, and customer language at the moment they are made, not in a quarterly retro. The principle is the same one IBM ran on the fab line. The reorg-resistant version of institutional knowledge is the one that lives in a file, not in the head of the person who left.

Connect the systems your team runs: The moat is not in the AI tool. The moat is in how deeply the AI tool reaches into the sheet and the inbox, the CRM, and the analytics layer. An AI tool that cannot read your content calendar is a chatbot. The same tool, hooked into the calendar with MCPs and pushing drafts straight into the social queue, becomes a marketing operator.

Ninety days is enough to ship the first round of all four. The second round is when the compounding starts. If you want help mapping the operator stack for a B2B marketing function, that is the work TinyTechGuides runs with marketing teams every week.

The moat eventually dries up

Brand, trust, and data flywheels are real moats. Operator depth is also a real moat. Brand and trust take years and millions. Operator depth takes a quarter and curated hours.

The marketing functions building operator depth in 2026 are producing differentiated work faster than the brand-only camp can match. They are not winning forever. Eventually, the practice gets commoditized too. The toolkit gets standardized. The compounding flattens. Every moat in the history of marketing has had a ceiling, and operator depth will be no exception.

For the next year or so, it is the moat available to anyone willing to do the work. The window is open. The AI tools are sitting on the shelf. The discipline is the part nobody else is willing to copy yet.

If you want help mapping your function’s operator stack, that is the conversation. Until then, write the CLAUDE.md. The tools are already on the shelf.

Need help with your PMM strategy or compounding your knowledge? Schedule a consultation.

Share

Leave a comment

Frequently asked questions

What is a marketing moat in 2026?

A marketing moat is a competitive advantage that compounds over time and cannot be quickly copied. The 2026 conversation has converged on brand, trust, and data flywheels. Each takes years and millions to build. Operator depth, the codified workflows and integrated tools that determine how fast a marketing team produces work, has joined the list because the instruments to build it landed in 2024. Hamilton Helmer would call it Process Power, applied to the marketing function for the first time.

How does operator depth differ from a tech stack?

A tech stack is the set of AI and martech tools your team has bought. Operator depth is the codified practice built on top of those tools. It includes named workflows, project rulebooks, and accumulated memory, plus MCP integrations into the systems where work happens. Two marketing teams can run identical stacks and produce different outputs because one has codified its operating practice and the other has not. The tools commoditize. The codification compounds.

Where do brand and trust fit if operator depth is the moat?

Brand, trust, and data flywheels are still real moats. They run on different time cycles. A brand takes five years and eight figures of paid spend to build. Trust requires a multi-year arc of consistent shipping and customer outcomes. Operator depth compounds in ninety days on a budget of curated hours. A CMO working on differentiation in 2026 cannot wait until 2031 to catch incumbents. Operator depth is the input that makes the consensus moats reachable inside a single planning cycle.

Can a small team really build operator depth in 90 days?

Yes, and a one-person marketing function often moves faster than a twelve-person team because the codification work is sequential. The first round is mechanical. Inventory the recurring workflows and write the project rulebook. Then capture decisions in memory as they happen and connect the AI tools to the systems your team runs every day. Ninety days is enough time for the first pass. The compounding starts in the second, when the workflows sharpen, and the memory file grows past anyone’s recall.

How does this connect to Hamilton Helmer’s 7 Powers?

Helmer’s 7 Powers (2016) lists seven sources of durable competitive advantage. Six are company-level decisions that a CMO cannot act on from a marketing seat. Process Power is the seventh, and the one that a marketing function can build on its own. Helmer’s canonical example is the Toyota Production System. Operator depth is the same logic applied to a marketing line. Skills and project rulebooks codify the standard work, and memory plus MCPs keep the codification durable across reorgs and integrated with the systems the team runs every day.

What’s the first thing to build?

Start with the project rulebook, the file that holds your team’s operating rules in one place. CLAUDE.md or your tool’s equivalent system-prompt file takes a weekend to write and pays back the next time you ship a piece of work. Capture what your team does every time, what it never does, and the brand voice every piece carries. Add the competitors off-limits to cite. Skills, memory, and MCPs come after. Without the rulebook to anchor them, those three layers drift away from the brand.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Software Equity Group. “2026 State of SaaS M&A: Buyers’ Perspectives.” Software Equity Group, 2026. https://softwareequity.com/research/saas-ma-buyers-perspectives.

^[2]Kellogg, Dave. “Kellblog Predictions for 2026.” Kellblog, January 22, 2026. https://kellblog.com/2026/01/22/kellblog-predictions-for-2026/.

^[3]CTI Digital. “2026 Marketing Trends: Brand as the Deepest Moat.” CTI Digital, 2026. https://www.ctidigital.com/insights/2026-trends-brand-as-the-deepest-moat/.

^[4]Cen, Steven. “AI Killed the Feature Moat. Here’s What Actually Defends Your SaaS Company in 2026.” Medium, 2026. https://medium.com/@cenrunzhe/ai-killed-the-feature-moat-heres-what-actually-defends-your-saas-company-in-2026-9a5d3d20973b.

^[5]Helmer, Hamilton. 7 Powers: The Foundations of Business Strategy. Deep Strategy LLC, 2016. https://www.7powers.com/.

^[6]Helmer, Hamilton. 7 Powers: The Foundations of Business Strategy. Deep Strategy LLC, 2016. https://www.7powers.com/.

Kate Strachnyi has a habit of calling people out on LinkedIn. When she spots a post that’s been run through an AI rewriter, she’ll send the person a direct message. “Hey, I could tell you used AI,” she’ll say. The usual response is some version of “but these are my thoughts,” and her advice back is to keep them that way and stop running them through the machine. She calls that keeping your content “non-GMO,” unmodified, and authentic.

It’s a funny line, and it also describes her entire business model. Kate is the founder of DATAcated, a media company that partners with brands in data, analytics, and AI to reach their audiences through real content creators and thought leaders. In a market where AI-generated posts are flooding every feed, and ironically, LinkedIn itself has a “rewrite with AI” button baked into the platform, Kate is making the opposite bet. She’s building a business around real people with real expertise and real opinions.

On Episode 38 of the Data Faces Podcast, I sat down with Kate to talk about how she built DATAcated from a one-person experiment into an influencer agency with 40+ creators, why she’s doubling down on authenticity as AI content takes over, and what happens to expertise itself when the humans who hold it stop doing the work.

“Keep it non-GMO. Don’t modify your content, just leave it as is.”

— Kate Strachnyi, Founder, DATAcated

About Kate Strachnyi

Kate Strachnyi is the founder of DATAcated. She started her career in finance and risk management consulting before pivoting to data visualization 12 years ago. Kate has since written five books, established one of the most connected networks of data and AI professionals in the industry, and built DATAcated into a full agency, with 40+ influencers, speakers, and subject-matter experts through her DATAcated Plus program. She is a LinkedIn Top Voice. In our conversation on Episode 38 of the Data Faces Podcast, we discuss:

- How Kate followed the revenue data from courses and books to a focused media business

- The DATAcated Plus model and how influencer campaigns work behind the scenes

- Why she’s shifting from “Kate = DATAcated” to an agency brand

- The flood of AI-generated content on LinkedIn and her “non-GMO” content philosophy

- What happens to expertise when today’s subject matter experts retire and AI fills the void

Following the data from finance to media

Kate’s path to running a media company started with a practical constraint. She was working in financial services risk management at a large consulting firm, traveling Monday through Thursday, and expecting her first child. She spent eight months searching for any role that would let her work remotely, well before remote work was mainstream. Someone eventually pointed her toward a data role that involved Tableau, visualization, and “creating pretty pictures.” She took it and fell in love with data storytelling.

“I am a data person, right? So I would look at the numbers and see what is driving more revenue and what is allowing me to have more time to myself.”

— Kate Strachnyi, Founder, DATAcated

What followed was a period of deliberate experimentation. Kate wrote books and launched an academy. She created courses, ran her own conferences, and built a community called DATAcated Circle. As a business of one with nobody to stop her, she could try anything, and she tracked the results. The revenue data and the work she enjoyed most both pointed to media and content creation. That’s where DATAcated lives today.

Building DATAcated Plus — from personal brand to influencer agency

When Kate started getting more client work than she could handle, she brought in the content creators she already knew. DATAcated Plus was born.

The program now spans data, analytics, and AI, with the agentic AI space growing fastest. When a client comes to Kate with a product launch, an event, or a brand awareness campaign, she matches them with creators based on their audience, expertise, and the success metrics the client is targeting. She reviews every piece of content before it goes to the client for approval. The day-to-day project management runs underneath, with UTM links, calls to action, and timelines coordinated across every moving piece. None of the work is glamorous, but it’s what keeps the whole operation running.

The DATAcated Plus roster also includes speakers and experts that companies can hire for keynotes and panels, as well as webinars and thought leadership papers. Kate has placed people across major industry events, including multiple years at the Gartner Data & Analytics Summit, where her team creates what she calls “FOMO-inducing content” on-site.

Scott Taylor, the Data Whisperer, joined Kate at last year’s Gartner event to co-lead a sold-out breakfast session on personal branding for data leaders. One question from the audience: “I want to post, but my company said no.” Kate’s advice was to follow your company’s rules but find the leeway. Talk about your perspective on industry topics rather than your specific projects or tools. When there’s no leeway at all, I suggested it might be time to find a company that sees your personal brand as an asset rather than a risk.

At Big Data London, a group of DATAcated Plus creators went to a tattoo parlor and got fake data tattoos for a video so convincing that Kate’s neighbors congratulated her on the new ink.

“People will unfollow instantly if we just keep promoting things. It’s more of letting my audience know about here’s a product that exists, and here’s what it does, in case you need it.”

— Kate Strachnyi, Founder, DATAcated

That creative range is what separates the DATAcated model from a traditional analyst engagement. Analyst firms produce authoritative research with independent evaluations, and content creators bring flexibility, personality, and a wider range of formats from short-form video to live event coverage.^[1] Kate’s crew has done cooking shows to explain data governance and built sandcastles to illustrate strong data foundations. The content sticks with audiences, and brands gain reach from creators who have built genuine trust over years of showing up with their own voices.

More recently, Kate has changed how she positions the company itself. The old DATAcated media kit led with a big photo of Kate and a rundown of what she could do. The new version leads with the influencer roster. She’s deliberately moving from “Kate equals DATAcated” to an agency brand that doesn’t depend on her being in every room. Some clients still ask for Kate specifically, and she’s learning to redirect them toward creators who are a better fit. “It’s nice to be wanted,” she told me, “but it doesn’t scale.”

The authenticity bet in an AI-saturated feed

Kate is leaning harder into genuine human voices at the exact moment AI-generated content is flooding LinkedIn and every other platform. LinkedIn added a “rewrite with AI” button to the post editor, while its own users complain that AI-generated slop is taking over their feeds. Kate’s view is that if you know a person well enough, you can spot AI-written content immediately. The vocabulary is off, the phrasing is too polished, and the voice sounds like everyone else’s. She calls people out on it, and she expects the same standard from her DATAcated Plus creators. For Kate, authentic content means it was written by the credited human, reflects their expertise and opinions, and hasn’t been reprocessed through an AI rewriter.

That doesn’t mean she’s anti-AI. Kate recently spent an entire day automating her invoicing process in Claude Code. The task itself takes two minutes, but she never has to do it by hand again. She and I compared notes about automating YouTube uploads, scheduling content, and eliminating the repetitive copy-paste work that eats up a solopreneur’s day. She draws the line between back-office operations and audience-facing content. AI can handle the invoices. It should not rewrite your LinkedIn posts.

“What are we going to do 20 years from now, when we don’t have those subject matter experts? They’re retired or not working anymore. Because if you don’t work with this stuff, whatever that might be in the medical field, in construction, how are you going to fact-check it?”

— Kate Strachnyi, Founder, DATAcated

Part of our discussion focused on a question Kate had heard at a recent event. Right now, subject matter experts can look at AI-generated output and spot what’s wrong because they’ve spent decades doing the work firsthand. But what happens in 20 years, when those experts have retired? If the next generation learns from AI output instead of from direct experience, the ability to verify and correct that output disappears.

The humans who make genuine content valuable are also the humans who keep AI honest. Kate sees her business as part of the answer. Invest in real experts now, amplify their voices, and make sure the knowledge doesn’t evaporate into a feedback loop of machine-generated content. The window to establish yourself as a real authority, someone whose voice carries weight because it’s grounded in lived experience, won’t stay open forever.

I started the Data Faces Podcast to have real conversations with the people doing the work. The messy, honest, sometimes funny exchanges that you can only get from humans who have opinions and aren’t afraid to share them. Kate’s business is built on the same conviction. In a world filling up with synthetic content, she’s betting that real voices will only become more valuable. When I asked her about deepfakes and AI versions of herself, her answer was four words.

“I plan to remain authentic.”

— Kate Strachnyi, Founder, DATAcated

Listen to the full conversation with Kate Strachnyi on the Data Faces Podcast.

Based on insights from Kate Strachnyi, Founder at DATAcated, featured on the Data Faces Podcast.

Leave a comment

Frequently asked questions

What is DATAcated, and what does the company do?

DATAcated is a media company founded by Kate Strachnyi, author of five books, including ColorWise and Journey to Data Scientist, that helps brands in data, analytics, and AI reach their target audiences through authentic content creators and thought leaders. The company runs the DATAcated Plus program, a roster of 40+ influencers, speakers, and subject-matter experts who create thought-leadership content, amplify product launches, and represent brands at industry events such as the Gartner Data & Analytics Summit. DATAcated operates as an agency that matches creators to client campaigns based on audience fit and success metrics.

What is the DATAcated Plus program?

DATAcated Plus is Kate Strachnyi’s influencer and speaker program for the data and AI industry. It includes content creators, speakers, and subject matter experts across data analytics, AI, and agentic AI. Companies hire DATAcated Plus members for brand awareness campaigns, event coverage, webinars, thought leadership papers, and on-site content creation. Kate manages the program by vetting creators for authenticity, reviewing all content before client approval, and coordinating timelines and deliverables across campaigns.

How does influencer marketing differ from analyst relations in B2B tech?

Analyst firms like Gartner and Forrester produce authoritative research and independent evaluations. Influencer content creators offer more creative flexibility and can be guided toward specific messaging for a campaign. Kate Strachnyi’s DATAcated Plus creators have done cooking shows to explain data governance and built sandcastles to illustrate data foundations. They’ve also produced viral video content at industry events. Both approaches build credibility with B2B audiences, and many companies now use influencers and analysts together at the same events.

What does “non-GMO content” mean in the context of AI and social media?

“Non-GMO content” is Kate Strachnyi’s phrase for content that hasn’t been run through an AI rewriter. Just as non-GMO food is unmodified, non-GMO content preserves the author’s original voice and phrasing. Kate advocates for this approach because AI-rewritten posts lose the personality and authenticity that make content creators valuable to their audiences. She actively calls out AI-washed posts on LinkedIn and holds her DATAcated Plus creators to the same standard.

Will AI replace subject matter experts in data and AI content?

Kate Strachnyi raises a concern about long-term expertise. Today’s subject matter experts can spot errors in AI-generated content because they have decades of hands-on experience. In 20 years, when those experts have retired, the ability to fact-check and verify AI output may disappear if the next generation learns from AI-generated content rather than direct experience. Kate’s business model is built around investing in real human experts now and amplifying their voices before that institutional knowledge erodes.

Podcast highlights

[0:05] Kate’s background and what DATAcated does

[2:10] Pre-finance Kate: what she wanted to be before data found her

[3:05] The career pivot from risk management consulting to data visualization

[5:03] How DATAcated evolved from training and books to a focused media company

[7:27] How the influencer model works behind the scenes

[9:33] Automating business operations with Claude Code

[11:01] Walking the line between brand amplification and spam

[14:11] The fake tattoo story from Big Data London

[15:03] How DATAcated Plus compares to analyst firm engagements

[17:14] The sold-out personal branding session at Gartner with Scott Taylor

[22:15] Shifting from “Kate = DATAcated” to an agency brand

[24:02] What works on LinkedIn now vs. five years ago

[27:01] AI-generated content flooding feeds and the “non-GMO” philosophy

[29:04] The 20-year question: who fact-checks AI when the experts retire?

[30:20] Deep fake Dave and why Kate plans to remain authentic

[31:24] Why Kate hasn’t hired a team and is betting on AI for operations

[33:57] Does AI make you more productive or just busier?

[36:19] Where to find Kate and DATAcated

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Sweenor, David. “Stop Writing AI Content That Sounds Like Everyone Else’s.” TinyTechGuides, February 7, 2025. https://insights.tinytechguides.com/p/stop-writing-ai-content-that-sounds

Skip the Claude mode debate

Most marketing teams I’ve worked with this past year are running similar setups. ChatGPT or Gemini lives in a permanently open browser tab, and the team’s saved prompts live in a Slack channel or Google doc that they copy and paste from. That’s a great way to get started, but as usage matures, the team needs more coordination and better tooling. Teams have seen all of the hoopla around Claude Code, and many of my clients are moving in that direction. Inevitably, they always end up asking the same question. How do I get started with Claude and which mode (Chat, Cowork, or Code) should I be using?

Take a gander around Substack and you’ll see an endless debate about which mode to use when. Chat for quick questions, Cowork to help connect your files and tools like Gmail, and Code for the technical bits. My recommendation is the same one I follow myself. Use only Claude Code. That’s it.

Remember the movie The Matrix? This is the same red pill / blue pill choice that Neo had. The browser tab is the blue pill. It’s easy and familiar, every conversation is a fresh start, and nothing compounds. Claude Code is the red pill. The editor is unfamiliar for a day. After that, you can’t unsee what’s on the other side and there’s no turning back.

The install takes 15 minutes, the vocabulary another 10. This post covers both, plus the case for skipping the debate.

A tool is not a stack

Installing Claude doesn’t change the AI itself. The model in your browser tab and the model in Claude Code are the same Anthropic model. But, the wrapping and context around the model differ. The additional context you build inside Claude Code turns the AI from a tool into a full-fledged marketing operating system. You can use it on your own, or, as your team scales, you can share context across your entire marketing or GTM function.

Browser-tab AI gives you the model and a chat window. That’s it. Memory exists, but it’s behind a separate settings screen, and you update it manually. The memory itself may leak across contexts you didn’t ask it to span. When you advise multiple clients or work on anything sensitive like product roadmaps, acquisitions, and launch plans, that can be problematic to say the least. The conversation you had about your upcoming product launch is one ambient prompt away from showing up in your conversation or content that’s being pushed to public channels before you’re ready. Consultants don’t want info from Client A leaking into Client B. You have better control over this in Claude Code.

Chat-only setups don’t share very well. Your colleague can’t inherit the prompts and context you’ve built up. Every team conversation is like a Saturday Night Live (SNL) cold open. Claude Code gives you the model plus four pieces that surround it that are transformative..

1. Memory you can scope to a project, so what Claude knows about Client A stays with Client A.

2. Skills you write once and reuse across the team with a slash command.

3. CLAUDE.md is a markdown file the team can read and edit, so the brand voice and deliverable formats are the same whether you’re drafting or your colleague is.

4. MCPs wire Claude into the tools you already use, like Gmail, Sheets, and Calendar. You don’t switch tabs or copy data into a chat. Claude reaches into the tools where the data lives.

Skills, CLAUDE.md, and MCPs all live in Code mode. Memory follows you across modes, and Code lets you scope it per project. That’s not coincidence. It’s why the rest of this post leans toward Cursor.

The 15-minute install

Four installs, about 15 minutes total. You can do this between meetings. I’ve run this install on my Mac and helped clients on Windows machines, and it doesn’t take long. To get started, follow these steps:

1. Start with a free GitHub account (2 minutes). This is where your team’s repo lives. Have the repo owner add you and accept the invite when it lands in your inbox.

2. Next, download Cursor and sign in with your work account (5 minutes). Cursor is the workspace where Claude Code runs. Defaults are fine and there is no need to buy a subscription.

3. Then install the Claude desktop app (3 minutes). You’ll mostly live in Code, but the desktop app is where you manage your account, and it pairs with Claude Code on the same login.

4. Finally, install Claude Code itself (5 minutes). Open Cursor’s terminal (View → Terminal) and paste the install command for your OS. On macOS or Linux, that’s curl -fsSL https://claude.ai/install.sh | bash. On Windows PowerShell, it’s irm https://claude.ai/install.ps1 | iex. Type claude and sign in when prompted. Claude Code requires a paid Anthropic plan (Pro, Max, Team, or Enterprise).

Here’s the checklist:

• Open Cursor and sign in

• Open the Claude desktop app and sign in

• Run claude in Cursor’s terminal without errors

• See your team’s repo on your GitHub account

When you get stuck, ask Claude. It can walk you through any install error, terminal command, or path issue you hit.

The three modes (and the one you need)

“Chat answers. Cowork connects to your tools. Code does all of that and more.”

— David Sweenor, Founder/CEO, TinyTechGuides

You’ll see three modes in the docs. Here’s what each one does, then what to do with that information.

• Chat: Back-and-forth conversation in a window. Best for quick questions, drafts, and brainstorming.

• Cowork: Claude plans and executes a task on your files, apps, and browser. Best for multi-step work that doesn’t live in a repo.

• Code: Terminal-based, runs inside a repo. Best for shared team docs, branded decks, and anything you want versioned.

Each mode has its own use case in the docs and its own corner of the marketing AI internet defending it. Last week, I saw a debate about whether spreadsheet research belongs in Cowork or Code. The week before, someone argued that brainstorming should never happen anywhere except Chat.

Each of the modes above is technically true, but none of them gets you to a marketing OS. Code does.

Code is where Skills, CLAUDE.md, and MCPs live. MCP stands for Model Context Protocol, the way Claude connects to other systems like Gmail, Asana, Canva, and Calendar. Code is where work compounds across sessions, where memory gets project-scoped, and where the team can see and edit what you made (via GitHub).

The most underrated benefit shows up when you need something none of those four pieces provide. An action-item tracker that pulls from your meeting notes. A status-report rollup across all your clients. Or a one-off Python script to clean up your Downloads folder. Code writes that on the fly. Browser-tab AI can’t write code that touches your files. Code can.

Teams matter too. When you ship work in Code to GitHub, your colleague clones the repo and inherits everything you’ve built. The Skills you wrote. The CLAUDE.md that defines the brand voice. The project-scoped memory rules. Browser-tab AI doesn’t ship. Your work stays on your account, where no one else can use it.

The Cursor interface is unfamiliar for the first day. The first 30 seconds are the worst. There’s a sidebar, a file tree, and a terminal pane. Push through it. After that, you’ve crossed the line, and you don’t go back, ever.

The four components that make Claude Code so powerful

Four components configure Claude. Three define how Claude works. The fourth wires Claude into the tools where your work already lives.

Memory is about you. Your role, your voice preferences, and how you like content formatted. Personal and portable across every project you work in. You set it once, and Claude carries it from session to session. After three months of corrections, the memory you’ve built is what makes Claude sound like you instead of a smart intern.

Skills are about the task. A Skill is a reusable recipe (instructions plus examples) that you call with a slash command. /write-blog. /generate-deck. /prep-podcast. Each one bundles the steps Claude should take for that task and the format of the output. For anyone already writing prompt workflows, Skills are how those workflows become team-shareable assets in your repo. Write a Skill once, and it works the same way every time you call it.

CLAUDE.md is about the project. It’s a markdown file in the project’s repo that Claude reads every time it works in that scope. It’s where you write down what’s true for this project. Brand voice. Audience. Deliverable formats. Drop one in your TTG repo, drop another in your client repo, and Claude switches voices without you reminding it.

MCPs wire Claude into the tools where your work already lives. Where Memory, Skills, and CLAUDE.md tell Claude how to work, MCPs tell Claude where to reach. Each MCP connects Claude to a specific tool, like Gmail, Sheets, Calendar, or Asana. You don’t have to copy data into a chat. Install one and Claude reaches into that tool. Install several and Claude works across your whole toolchain.

“Memory is about you. Skills are about the task. CLAUDE.md is about the project, and MCPs connect your apps.”

— David Sweenor, Founder/CEO, TinyTechGuides

A growing set of community patterns extends the same idea. A popular one is napkin.md, a per-repo runbook that Claude reads at session start and curates as you work. It catches mistakes once and stops repeating them. You commit it or you don’t, your call.

Your first move

You finished the install. Now what?

Open Cursor’s terminal, run claude, and have a five-minute conversation. Tell Claude who you are. Tell it what you do. Tell it the three things you reach for most often in a week. Save what you tell it as memory.

That’s the first move. Anything more is week-two work. Your first Skill conversion, your first CLAUDE.md, and your first MCP can wait. None of that matters in week one. Build the habit of opening Cursor, typing claude, and treating it like a colleague who started Monday.

Take the red pill

Fifteen minutes is the easy part. The IDE filters out most marketers. The marketing OS rewards the ones who push through.

Most marketers won’t cross the gate, which is why their AI output never compounds. You’re reading this, which means you’re closer than you think. Cross the gate. Take the red pill.

This post is the entry point for the Claude for Marketing Operators series on TinyTechGuides. The next eight posts walk through each layer of your marketing OS. The four-layer mental model. CLAUDE.md teardowns from real client projects. The MCPs that pull their weight, and the ones that don’t. What six months of memory looks like across multiple clients. How a solo consultant runs four engagements on one brain. Why your marketing OS becomes the moat your competitors can’t copy.

If you want each one as it drops, subscribe to the Marketing section on Substack.

Share

Leave a comment

Frequently asked questions

What is Claude Code?

Claude Code is the terminal-based mode of Anthropic’s Claude AI assistant, built for work that lives in a code repository. For marketers, it runs inside an editor like Cursor and lets you maintain a project rulebook (CLAUDE.md), reusable Skills, scoped memory, and connectors (MCPs) to tools like Gmail and Sheets. It’s also the only mode that can write Python on the fly to handle tasks none of those four pieces cover, like building an action-item tracker or a status-report rollup across clients.

How does Claude Code differ from Chat and Cowork?

Chat is back-and-forth conversation. Cowork lets Claude plan and execute multi-step tasks across your files, apps, and browser. Code runs in your terminal inside a code repository, so the work compounds across sessions through Skills, CLAUDE.md, scoped memory, and MCPs. Chat and Cowork are useful for narrow tasks. Code is where reusable assets live and where teams can share what one person built. For marketing operators using Claude in production, the recommendation is to skip the mode debate and run everything in Code.

Do I need to be a developer to use Claude Code?

No. Claude Code runs in a terminal, but you don’t need to write code to use it. You install it once, then talk to Claude in plain English. The terminal interface is unfamiliar for the first day, but most marketing tasks (drafting blog posts, building decks, summarizing client meetings, packaging prompts as reusable Skills) work the same way they do in a chat window. The benefit is that everything you do compounds across sessions and can be shared with your team through a repo.

Does Claude Code require a paid Anthropic plan?

Yes. Claude Code requires a Pro, Max, Team, or Enterprise plan from Anthropic. The free Claude.ai plan does not include Claude Code access. The desktop app, browser-based Claude.ai, Cursor, and GitHub are all free — only the Claude Code CLI itself needs a paid plan. You can install all four pieces (GitHub, Cursor, Claude desktop app, Claude Code) before deciding on a plan, since the install completes before authentication.

What’s the difference between memory and a CLAUDE.md file?

Memory is personal and portable across every project you work in: your role, voice preferences, and how you like content formatted. CLAUDE.md is project-specific. It’s a markdown file in a project’s repo that Claude reads every time it works in that scope, capturing brand voice, audience, deliverable formats, and anything else that’s true for that project. Memory follows you across modes; CLAUDE.md stays with the project. A solo consultant runs one memory and many CLAUDE.md files, one per client.

Where should I start if I want to use Claude for marketing work?

Install the four pieces (GitHub, Cursor, Claude desktop app, Claude Code) in about 15 minutes. Then open Cursor’s terminal, run claude, and have a five-minute conversation. Tell Claude who you are, what you do, and the three things you reach for most often in a week. Save what you tell it as memory. That’s the entire first move. Skill conversion, CLAUDE.md drafting, and MCP setup are all week-two work. Week one is opening the terminal and starting to use it.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

Four out of five organizations are deploying AI. Only one in five is hitting its ROI targets.^[1] I’ve been to the Gartner Data & Analytics Summit eight or nine times over the years, and I’m not sure if it’s déjà vu or a movie I’ve already seen.

When big data was en vogue, we saw the same thing. Everyone rushed to adopt the technology, even though few understood it or could articulate the returns. The pay was always “next year.” Big data had its 3 Vs of Volume, Velocity, and Variety. Some people stretched it to 7, because apparently three wasn’t enough to describe the problem. The AI era already has its own shorthand, and I’m calling them the 3 Ws – washing, wishing, and waiting.

I spent three days at the summit in Orlando this March, and the 3 Ws were everywhere. They explain why that 1-in-5 number is so low, and they’re the same symptoms that dragged big data down a decade ago.^[2]

Washing

The vendor side and the practitioner side might as well have been at two different conferences. On the expo floor, every booth was a native “AI agent” company. I couldn’t figure out what half of them did because the messaging was identical. Big data washing became AI washing, which became agent washing, and the label keeps rotating while the behavior underneath stays the same.

A Gartner analyst warned against this pattern in a session on AI agents in analytics and BI, recommending evidence-based evaluation with real-world scenarios instead of trusting vendor marketing claims.^[3] Good advice, but the expo floor didn’t get the memo. Booth copy sounded like it was written by the same LLM, and vendor presentations felt forced. The AI slop is feeding itself, and Gartner predicts a $58 billion market shakeup as GenAI and AI agents challenge mainstream productivity tools.^[4]

“Big data washing became AI washing, which became agent washing.”

— David Sweenor, Founder/CEO, TinyTechGuides

Wishing

The hallway conversations told a different story compared to the keynotes. Practitioners weren’t talking about what’s working, but about what isn’t working yet. Session after session pointed to 2027 and 2030 as the years when AI will finally achieve nirvana and deliver on its promises.^[5] The finish line keeps moving, and big data had the same timeline problem. The returns were always coming “next year,” just around the corner, right after you got your data sh*t in order.

Six out of ten IT leaders are worried about AI agent cost overruns, but only two out of ten D&A leaders share that concern.^[6] That disconnect should worry everyone. The people closest to the data aren’t worried about the costs, and the people paying the bills are.

“The returns were always coming next year, just around the corner.”

— David Sweenor, Founder/CEO, TinyTechGuides

We might be one bad quarter away from CFOs redirecting AI budgets to other mission-critical priorities, or worse yet, replacing agents with actual employees. When results don’t materialize, the panacea that is AI dissipates, which is what happened to big data. The enthusiasm evaporated the moment CFOs started asking tough questions about returns, and while the technology survived, most of the budgets didn’t.^[7]

Waiting

The “vendor conference” sold AI-first futures, while the practitioners who need to make all of this work understood that the foundations aren’t there. Everyone is waiting, and the waiting comes in three flavors.

“Organizations are building AI on top of foundations they don’t trust.”

— David Sweenor, Founder/CEO, TinyTechGuides

Waiting on the data

Most catalog and context layer vendors are still built for structured, tabular data. Today’s AI strategies depend on documents, images, and code, which current tooling wasn’t designed to handle, and no one seems to be addressing. Meanwhile, the conversation has moved to metadata, semantic layers, and context layers as the connective tissue that makes AI work. As Scott Taylor puts it, context is the new oil.^[8] By 2027, Gartner expects 40% of IT spending on data management to target multistructured data, and AI data readiness spending will increase sevenfold from 2025 to 2029.^[9] The money is apparently coming. Who’ll build the plumbing? Agents or people?

Waiting on governance

Everyone is pro-governance, and for most companies in the US, that’s about as far as they’ll go. AI governance, the kind that covers model risk management, bias testing, usage controls, and agentic oversight, is barely on the radar.^[10] Only 14% of IT leaders said they were confident their data and content assets are secured and governed.^[11] Organizations are building AI on top of foundations they don’t trust.

Waiting on people

Gartner reports that only 6% of D&A and AI leaders consider themselves fully AI-ready when it comes to people, skills, and change management.^[12] I’m not sure I agree with that number. People are using AI all over the place, whether their organizations are ready or not. The formal readiness programs haven’t caught up yet, but informal adoption has already happened. Change management alone can require up to twice the effort of the implementation itself.^[13] Scott Brinker calls this pattern Martec’s Law, where technology changes exponentially while organizations change logarithmically.^[14] That’s why most companies are still staring at the empty shelf long after the technology has moved three aisles over.

Big data had all three of these problems. Organizations bought the tools before fixing the plumbing, skipped governance, and underinvested in people. The AI era is repeating the same mistakes.

Same plot, different city

The Hangover Part II was essentially the same movie in a different city, with a new hotel and another bachelor party leading to the same bad decisions. That’s what this feels like.

The reasons four out of five organizations aren’t hitting their AI ROI targets are the same reasons big data underdelivered. Only the buzzword changed. The organizations that closed the distance last time invested in the boring stuff, the blocking and tackling.^[15] They focused on governance before it was fashionable, fixed data quality before the next platform purchase, and developed people instead of cutting headcount. None of it was exciting, but all of it worked.

The technology works, and the big data technology worked too. What didn’t deliver was everything around it,^[16] which is where the washing, wishing, and waiting did their damage. No new generation of AI agents will fix that for you.

Share

Leave a comment

Frequently asked questions

What are the 3 Ws of AI?

The 3 Ws of AI are Washing, Wishing, and Waiting. Washing is vendors rebranding existing products as “AI agents” without real AI underneath. Wishing is that practitioners are deferring AI ROI to 2027 and 2030, while only one in five organizations hits current ROI targets. Waiting is the shortfall in data readiness, governance, and change management. Together, the 3 Ws explain why most AI investments fall short.

What is AI washing?

AI washing is the practice of rebranding existing products as “AI” or “AI agent” without meaningful AI underneath. At Gartner D&A Summit 2026, nearly every expo booth marketed “AI agent” products with near-identical messaging. The pattern mirrors big data washing from a decade ago. The label keeps rotating while the behavior underneath stays the same.

Why are so few AI deployments hitting their ROI targets?

The 2026 Gartner CIO Survey shows that four in five organizations are deploying AI, but only one in five hits ROI targets. The failure pattern repeats big data’s mistakes. Organizations bought tools before fixing data foundations, skipped governance (only 14% of IT leaders feel confident), and underinvested in people. The technology works; the execution around it keeps failing.

Why are IT and D&A leaders divided on AI cost concerns?

At the 2026 Gartner D&A opening keynote, six out of ten IT leaders said they were worried about AI agent cost overruns, but only two out of ten D&A leaders shared that concern. The people closest to the data aren’t worried about costs, and the people paying the bills are. AI budgets may come under CFO scrutiny before D&A teams expect it.

Where should D&A leaders focus to improve AI ROI?

Focus on the three foundations most organizations are waiting on: data readiness, governance, and people. Gartner expects 40% of IT spending on data management to target multistructured data by 2027, with AI data readiness spending growing sevenfold from 2025 to 2029. Only 14% of IT leaders are confident in governance, and only 6% of D&A leaders are AI-ready on people and change management.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]2026 Gartner CIO Survey, n=2,437.

^[2]Sweenor, David. “Beyond the AI Hype: What 20% of Companies Get Right.” TinyTechGuides, February 11, 2025. https://tinytechguides.com/blog/beyond-the-ai-hype-what-20-of-companies-get-right/.

^[3]Macari, Edgar. “Activating AI Agents in Analytics and BI Platforms.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[4]Sallam, Rita. “Top Data and Analytics Predictions for 2026.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL. See also Gartner, “Gartner Announces Top Predictions for Data and Analytics in 2026,” press release, March 11, 2026, https://www.gartner.com/en/newsroom/press-releases/2026-03-11-gartner-afromnnounces-top-predictions-for-data-and-analytics-in-2026.

^[5]Sallam, Rita. “Top Data and Analytics Predictions for 2026.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL. See also Gartner, “Gartner Announces Top Predictions for Data and Analytics in 2026,” press release, March 11, 2026, https://www.gartner.com/en/newsroom/press-releases/2026-03-11-gartner-announces-top-predictions-for-data-and-analytics-in-2026.

^[6]Ronthal, Adam, and Georgia O’Callaghan. “Opening Keynote: Navigate AI on Your Data & Analytics Journey to Value.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[7]Sweenor, David. “The AI-Powered CFO: Why Finance Must Shift from Control to Cognition.” TinyTechGuides, March 11, 2025. https://tinytechguides.com/blog/the-ai-powered-cfo-why-finance-must-shift-from-control-to-cognition/.

^[8]Sweenor, David. “Truth Before Meaning — The Three-Word Fix for Data Management.” TinyTechGuides, April 7, 2026. https://tinytechguides.com/blog/truth-before-meaning-the-three-word-fix-for-data-management/.

^[9]Showell, Nina. “Unstructured Data Is the Missing Ingredient to Prepare AI-Ready Data.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[10]Sweenor, David. “Why Bad AI Governance Kills 95% of Enterprise Projects Before Production.” TinyTechGuides, September 9, 2025. https://tinytechguides.com/blog/why-bad-ai-governance-kills-95-percent-enterprise-projects/.

^[11]2025 Gartner GenAI Enterprise Survey, n=360.

^[12]2025 Gartner Survey, n=353.

^[13]Ronthal, Adam, and Georgia O’Callaghan. “Opening Keynote: Navigate AI on Your Data & Analytics Journey to Value.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[14]Brinker, Scott. “Martec’s Law: Technology Changes Exponentially, Organizations Change Logarithmically.” Chiefmartec, June 13, 2013. https://chiefmartec.com/2013/06/martecs-law-technology-changes-exponentially-organizations-change-logarithmically/.

^[15]Sweenor, David. “Why 80% of AI Projects Fail (And the Three Boring Decisions That Save the Other 20%).” TinyTechGuides, October 21, 2025. https://tinytechguides.com/blog/why-80-of-ai-projects-fail-and-the-three-boring-decisions-that-save-the-other-20/.

^[16]Sweenor, David. “The $40B Reason Enterprise AI Projects Fail: It’s Not the Tech.” TinyTechGuides, September 6, 2025. https://tinytechguides.com/blog/the-40b-reason-enterprise-ai-projects-fail-its-not-the-tech/.

For 25 years, data quality has been everyone’s problem and nobody’s priority. For some, it was an IT problem, and for others, it was a business problem. But most of the time, fixing it at scale was largely ignored. What would you do if a number in the spreadsheet looked off? You’d fix it and move on with your day. The same with questionable metrics on dashboards. We’ve been able to tuck and hide the cost of bad data in a manual world for a while now. Since the pace of business was slower, there were no real consequences for getting it wrong.

Those ways of old change when you hand autonomy to an AI agent. An agent doesn’t pause to gut-check a suspicious number, it doesn’t really care. It takes the data at face value, makes a decision, feeds that decision into the next step, and keeps going. You might be six or seven steps down the line before anyone realizes the foundation was wrong. And by then, the damage compounds in ways that a quick spreadsheet fix can’t undo.

I sat down with Brendan Grady, EVP and General Manager of Analytics and AI at Qlik, at Qlik Connect 2026 in Orlando to discuss why the stakes around data quality have changed, where enterprise-agentic adoption stands today, and what data professionals should be thinking about.

“In today’s world where there may be an agent running around using said data and getting it wrong, the consequences of getting it wrong are going to be catastrophic.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

About Brendan Grady

Brendan Grady is EVP and General Manager of the Analytics and AI Business Unit at Qlik, where he leads product management, product design, R&D, and go-to-market strategy for the company’s data integration, quality, and analytics platform. Before Qlik, he held senior GTM roles at IBM, where he led worldwide digital sales for Watson Analytics and managed the Cognos portfolio. He joined Qlik seven years ago after repeatedly losing deals to its analytics engine, and decided to find out why. And well before all of that, he delivered the Sound of Music tour in Salzburg, Austria, over 300 times.

In this episode, we discuss:

- Why data quality was never fixed and why that matters now

- Where enterprise agentic AI adoption actually stands

- Trust scores and the problem with feeding spreadsheets to LLMs

- The shift from dashboards to decision intelligence

- Open standards, MCP, and why there’s no “One Ring to rule them all”

- Advice for data professionals navigating the AI transition

The consequence management problem

Grady framed the data quality conversation in a way I hadn’t heard before. He called it consequence management. For decades, organizations tolerated bad data because the consequences of getting it wrong were manageable. A field was incorrect in a report? Someone caught it, fixed it, and everyone moved on, knowing there would be another fire drill tomorrow. The recovery cost was low enough that nobody prioritized prevention, and if they did, they rarely had the organizational backing to make any meaningful change.

“Is it IT’s job? Is it the business’s job? Is it both, or is it nobody’s job? For most companies, it’s been nobody’s job.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

BARC’s research confirms this pattern. As Shawn Rogers discussed on the Data Faces Podcast, data quality remains the top challenge for organizations trying to mature their analytics and AI capabilities.^[1] That organizational ambiguity persisted because the stakes allowed it. He pointed to real examples. A major airline took a significant hit to its market cap because its sentiment data was wrong and decisions were made on flawed analysis. Two decades ago, a single field in a spreadsheet contributed to a financial crisis that rippled through an entire market. These weren’t hypothetical scenarios. They happened because nobody owned the problem and the systems in place couldn’t detect the errors before they cascaded.

In the agentic era, the failure mode is different. A human looking at a dashboard might notice something feels off and investigate. An agent won’t. It will take the data, reason through it, make a decision, and pass that decision to the next agent in the chain; often without any confidence bounds or trust scores.

The point isn’t that agents are dangerous. The point is that autonomous systems need trusted data underneath them before they’re given the authority to act. Without that bedrock, every step an agent takes amplifies whatever error was baked into the starting point. As practitioners, we know this, why hasn’t this been fixed yet?

“Prior to stage zero”

I asked Brendan where enterprise agentic adoption actually stands. His answer was honest. “What’s prior to stage zero?” he said. “I mean, there are customers that are trying things out there, surely. But from a large-scale production standpoint, we’re in the early days.”

Customers are experimenting with low-risk use cases. They’re testing agents in controlled environments where a mistake won’t damage the business. But production-grade agents making real decisions in real business processes? That’s rare. And the blocker, according to Brendan, isn’t the technology. It’s the data.

Gartner projects that by 2027, 70% of organizations will adopt modern data quality solutions to support AI adoption and digital business initiatives.^[2] That projection tells you where the market is today. If 70% will need to adopt these solutions by 2027, most organizations don’t have them yet. The ambition around agentic AI is running well ahead of the data infrastructure required to support it. Shane Murray made a similar argument on the Data Faces Podcast earlier this year, noting that actionable data strategies beat endless planning when it comes to AI readiness.^[3]

Brendan also raised a practical question that every data leader should be asking. The LLM landscape is shifting constantly. Six months ago it was OpenAI. Today, Claude is gaining traction. Tomorrow the market may have moved on to something new. His advice was to work with vendors that approach this from an open standards perspective, supporting multiple LLMs rather than forcing a single choice. The technology will keep changing, but the data underneath it is what has to hold steady.

“The internet took 10 years, 20 years, 30 years to get going. We’re a year and a half in.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

Trust as the missing layer

One of the more revealing moments in our conversation came when Brendan talked about what happens when you feed structured data into an LLM. I’ve experienced this myself. You upload a spreadsheet, ask it to calculate something, and the answer comes back looking polished and confident. The formatting is clean, the language is professional, and unbeknownst to you, the numbers are wrong.

“It’s really pretty, right? The answer is amazing. Looks great. Totally BS. And the next thing you know, you’re showing up to the board with all incorrect numbers.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

Qlik’s Trust Score for AI is designed to give decision-makers a quantifiable measure of whether their data is valid, fresh, and representative before it reaches an agent or an LLM.^[4] Instead of hoping your data is accurate, you can see a score that tells you it’s 90% trustworthy or 80% or something that should give you pause.

The other piece Brendan emphasized was intent detection. When someone asks a question of an LLM, the literal question and the actual intent are often different things. I ran into this recently when I asked an AI assistant to analyze several websites. It came back with a confident analysis, but when I pressed it, it admitted it had never actually visited the sites. Qlik is investing in understanding what the user is really trying to accomplish so the system can route to the right data and the right engine rather than letting an LLM fabricate its way to an answer.

The combination of trust scores and intent detection reflects a broader principle. Before you give an agent the authority to act on data, you need to know that the data is sound and that the system understands what you’re actually asking. Qlik’s track record in this space is long. The company has been named a Leader in the Gartner Magic Quadrant for Augmented Data Quality Solutions for seven consecutive years, most recently in 2026.^[5]

“Dashboards are dead. Long live dashboards.”

When Brendan declared that dashboards are dead, I thought I had a scoop and I made sure the audience heard it. He laughed and then walked it back with the nuance that matters. Dashboards as a destination are going away, but the data inside them and the decisions they inform are more important than ever.

Brendan described how his own workflow has changed. He used to ask his analytics tools for information about business performance. Now he asks a different question. “Tell me about my business and what you think I should do.” That shift from information retrieval to decision recommendation is what Qlik means by decision intelligence, and it’s powered by two things working together.

The first is Qlik’s analytics engine, which finds associations and relationships in data that other approaches miss. Instead of running a predefined query to answer a specific question, the engine surfaces connections you didn’t know existed. Brendan called these the unknown unknowns. In an agentic context, that capability becomes even more valuable because it allows agents to explore paths and relationships that a standard SQL query would never surface.

“In the agentic world, we’re serving this up to help agents understand that there’s a relationship here that you need to go explore before you take action. That is extremely powerful.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

The second is openness. Qlik launched its MCP server in February 2026, implementing the open Model Context Protocol standard to let third-party AI assistants access Qlik’s analytical capabilities with governance built in.^[6] “There’s never going to be One Ring to rule them all,” he said. People want to work in the tools they’re comfortable with, whether that’s Claude, Gemini, ChatGPT, or something that doesn’t exist yet. The bet is paying off. Brendan shared that they’re already seeing roughly a 50/50 split between users accessing agentic capabilities through Qlik’s own interface and those coming in through MCP.

“Am I out of a job?”

Brendan closed our conversation with a story that we’ve all encountered. After demoing the ability to build an analytics application through Claude in 30 seconds at Qlik Connect, a customer approached him. This person had built his entire career writing code to create analytics applications across multiple platforms. His question was simple. “Am I out of a job?”

Brendan’s answer was no, but with an important caveat. The job will evolve. His advice to data professionals was to lean into what they already know better than anyone else: the data itself. Become the data product owner. Be the trusted guide as organizations navigate the agentic experience. The people who understand the data well enough to know its quirks and business context will be indispensable as agents take on more routine work.

This tracks with what Brendan’s team has seen internally. Qlik has developers who were already performing well, and AI tools have turned them into 10x contributors. The acceleration is happening at the top end, where strong performers are getting faster and producing better work. A preliminary MIT Media Lab study found that heavy reliance on AI assistants can lead to what researchers called “cognitive debt,” where users outsource critical thinking and lose the ability to recall and synthesize what they’ve produced.^[7] Brendan acknowledged this risk directly. He sees his own daughters, 19 and 24, defaulting to LLMs for answers, and he worries about critical thought eroding over time.

“Embrace these new technologies. It’s scary. But your job will evolve. Become that data product owner, become an expert in that data, and be that trusted guide as everybody’s going down the agentic experience.”

— Brendan Grady, EVP and GM of Analytics & AI, Qlik

The real opportunity for data professionals is to become the people who make sure agents are working with the right information in the right context. That’s a role no LLM can fill on its own. If you’re not sure where to start, audit the data your team’s AI tools depend on. If you can’t quantify how trustworthy that data is, that’s the first problem to solve.

Listen to the full conversation with Brendan Grady on the Data Faces Podcast.

Based on insights from Brendan Grady, EVP and GM of Analytics & AI at Qlik, featured on the Data Faces Podcast.

Share

Frequently asked questions

What is consequence management in the context of data quality?

Consequence management is the idea that data quality was never prioritized because the consequences of bad data were manageable. In a manual world, a wrong number in a spreadsheet could be corrected before it caused real damage. With AI agents making autonomous decisions across multiple steps, errors compound before anyone detects them. Consequence management explains why the stakes around data quality have shifted from recoverable inconvenience to potential business-level damage.

Where does enterprise adoption of agentic AI stand in 2026?

According to Brendan Grady, EVP of Analytics and AI at Qlik, enterprise agentic adoption is in its earliest stages. Customers are experimenting with low-risk use cases in controlled environments, but production-grade agents making real decisions in real business processes are rare. Data quality is the primary blocker. Gartner projects that by 2027, 70% of organizations will adopt modern data quality solutions to support AI initiatives.

What is Qlik’s Trust Score for AI?

Qlik’s Trust Score for AI is a quantifiable measure of whether data is valid, up to date, and representative before it reaches an AI agent or a large language model. It scores data across dimensions including diversity, timeliness, and accuracy, giving decision-makers visibility into data reliability rather than requiring them to take data quality on faith. Qlik has been named a Leader in the Gartner Magic Quadrant for Augmented Data Quality Solutions for seven consecutive years.

What does “dashboards are dead” mean?

Brendan Grady’s declaration that “dashboards are dead” refers to dashboards as a destination, not the data or insights within them. The traditional model of going to a dashboard to draw your own conclusions is being replaced by AI-powered interfaces that proactively recommend actions. Qlik calls this shift decision intelligence. Grady described his own workflow changing from “give me information about my business” to “tell me about my business and what you think I should do.”

What is the Qlik MCP server?

The Qlik MCP server implements the open Model Context Protocol, allowing third-party AI assistants such as Anthropic Claude, Google Gemini, and ChatGPT to access Qlik’s analytical capabilities, with built-in governance and audit trails. Launched in February 2026, it reflects Qlik’s bet on interoperability over platform lock-in. Grady reported that roughly 50% of users now access Qlik’s agentic capabilities through MCP rather than Qlik’s own interface.

What should data professionals do to prepare for the agentic AI era?

Brendan Grady advises data professionals to lean into what they already know best: the data itself. His recommendation is to become data product owners who serve as trusted guides as organizations adopt agentic AI. The people who understand data quality, business context, and organizational nuance will be indispensable because these capabilities are not ones AI agents can replicate on their own.

Podcast highlights

- [0:00] Introduction and welcome at Qlik Connect 2026

- [1:14] Brendan’s first job: Sound of Music tour guide in Salzburg

- [2:04] Lessons learned from the early analytics era

- [3:32] Why data quality has never been fixed

- [4:46] Consequence management in the agentic era

- [6:08] Where enterprise agentic adoption actually stands

- [7:46] Future-proofing against LLM shifts

- [8:24] The analytics engine and unknown unknowns

- [10:29] Structured vs. unstructured data convergence

- [12:04] Hallucinations and the trust problem

- [15:30] Decision intelligence and “dashboards are dead”

- [18:05] Brain outsourcing and the MIT cognitive debt study

- [21:57] MCP server and open standards

- [23:54] Key themes for Qlik in 2026: trust, context, flexibility

- [26:12] Advice for data professionals

- [28:15] Does AI expand the aperture for who can participate in analytics?

About David Sweenor

David Sweenor is the founder and host of the Data Faces Podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Sweenor, David. “Beyond the AI Hype: What 20% of Companies Get Right.” TinyTechGuides, February 11, 2025. https://tinytechguides.com/blog/beyond-the-ai-hype-what-20-of-companies-get-right/

^[2]Gartner. “Lack of AI-Ready Data Puts AI Projects at Risk.” Gartner Newsroom, February 26, 2025. https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk

^[3]Sweenor, David. “From ‘AI-Ready’ to AI Reality: Why Actionable Data Strategies Beat Endless Planning.” TinyTechGuides, June 3, 2025. https://tinytechguides.com/blog/from-ai-ready-to-ai-reality-shane-murray-on-data-trust-and-why-action-beats-planning/

^[4]Qlik. “Qlik Releases Trust Score for AI in Qlik Talend Cloud.” Qlik Press Release. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-releases-trust-score-for-ai-in-qlik-talend-cloud

^[5]Qlik. “Qlik Named a Leader in the 2026 Gartner Magic Quadrant for Augmented Data Quality Solutions.” Qlik Press Release, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-named-a-leader-in-the-2026-gartner-magic-quadrant-for-augmented-data-quality-solutions

^[6]Qlik. “Qlik Brings Agentic Analytics to General Availability and Launches MCP Server for Third-Party Assistants.” Qlik Press Release, February 10, 2026. https://www.qlik.com/us/news/company/press-room/press-releases/qlik-brings-agentic-analytics-to-general-availability-and-launches-mcp-server-for-third-party-assistants

^[7]MIT Media Lab. “Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Task.” MIT Media Lab, 2025. https://www.media.mit.edu/publications/your-brain-on-chatgpt/

On the Data Faces Podcast, I usually interview someone with a whole face. For this bonus segment, I made an exception.

Scott Taylor, the Data Whisperer, is known for his work in data management consulting and storytelling. He’s also the creator of Data Puppets, a satirical puppet series that uses humor to expose the enterprise data problems that executives resist hearing about directly. In Episode 35 of the Data Faces Podcast, Scott and I had a serious conversation about why data leaders keep losing the room and how storytelling wins it back. Then we let one of his puppet characters take the mic.

The character’s name is A-Eye. Not “an AI.” As the puppet put it, “I choose my own indefinite article. Personal branding is important.”

A-Eye had just returned from the Gartner Data & Analytics Summit. The report from the show floor was not subtle.

“AI was everywhere. Regular AI, gen AI, agentic AI, autonomous AI, in the loop AI, out of the loop AI, trapped in the loop AI. Some vendors were basing their whole future on it, and two years ago they couldn’t even spell AI.”

The agents impressed him most. “Agents writing code, agents reviewing code, deploying code, and then apologizing for the code. It’s a total system.”

I asked about data quality. A-Eye was unmoved. “They’ve been whining about data quality ever since there was data. If it was that important, would it have been solved by now?”

And governance? “AI is the Ozempic for data governance, baby. Your data never looked so good.”

The segment wrapped with A-Eye reworking Old MacDonald into a data anthem. I was asked to sing along. I did. I shouldn’t have.

Scott’s Data Puppets work because a puppet can say things that would come off as harsh from a human consultant. The CDO (Chief Dog Officer), IT Bee (who speaks only in buzzwords), and the Cat Sultant from Meow-kinsey have all become tools that data teams use in their own presentations to show leadership how the data team sounds to the business side. The satire lands because it’s uncomfortably accurate.

Watch the full conversation with Scott Taylor on the Data Faces Podcast.

Data leaders have been pitching “data quality” to executives for decades. For just as long, executives have nodded politely, approved a fraction of the requested budget, and moved on to whatever initiative sounds more exciting. Gartner estimates that poor data quality costs the average enterprise $12.9 to $15 million per year, yet data leaders still struggle to connect that cost to the language executives actually use.^[1]

“I can boil my entire data philosophy down to those three words: truth before meaning. You got to determine the truth in your data before you derive any kind of meaning out of it. It’s not chicken or egg here. This is egg and omelet.”

— Scott Taylor, Founder, MetaMeta Consulting

On Episode 34 of the Data Faces Podcast, I sat down with Scott Taylor to talk about why data management keeps getting sidelined and what data leaders can do about it. Scott has spent 30 years in the data space and now runs MetaMeta Consulting, where he helps organizations craft business-accessible narratives about data management. His central argument is that you have to establish truth in your data before you try to derive any meaning from it. Getting there requires storytelling, a skill that most data practitioners were never trained in.

About Scott Taylor

Scott Taylor is the founder of MetaMeta Consulting and is known across the data industry as “the Data Whisperer.” He has spent 30 years in the data space, including 25 years in corporate roles before becoming a full-time content creator, speaker, and consultant. Scott is also the creator of Data Puppets, a satirical puppet series that uses humor to expose common enterprise data problems, and the author of Telling Your Data Story. In our conversation on Episode 34 of the Data Faces Podcast, we discuss:

- Why “truth before meaning” is the foundational principle for every data initiative

- How data leaders can craft a one-sentence pitch that resonates with a skeptical CFO

- The 3V framework for data storytelling: Vocabulary, Voice, and Vision

- Why the vendor landscape at Gartner D&A looked “horrifyingly consistent”

- How Data Puppets uses satire to expose organizational dysfunction that executives resist hearing directly

Truth before meaning: egg and omelet, not chicken and egg

Every major trend in the data space follows the same cycle. A new technology arrives with enormous promise; organizations rush to adopt it; and at some point, someone in the room realizes that none of it works unless the underlying data is in good shape. Scott has watched this play out with big data, data mesh, data fabric, and now agentic AI over three decades. He frames the challenge with a phrase he has distilled into the fewest possible words: truth before meaning.

The “truth” part refers to the foundational work that most organizations know they need to do but struggle to prioritize: master data, reference data, metadata, MDM, data governance, and all the structural activities that ensure your data is curated, trusted, and fit for purpose. The “meaning” part is everything that gets the budget and the boardroom attention, from BI dashboards and AI models to analytics platforms and the latest agentic AI initiative.

Organizations consistently skip truth and jump straight to meaning. They invest in AI without verifying that their customer master is reliable. They build analytics on hierarchical structures that don’t hold up across departments.^[2] Research from IBM and MIT Sloan Management Review suggests that companies lose 15 to 25 percent of revenue because of poor data quality, and that cost only grows as organizations scale their AI investments without fixing the underlying data.^[3]

Scott illustrates the point with a supermarket example. Every time you take a product off a shelf and scan it at the register, that beep is a confirmation of truth. The system knows exactly what that product is, how it’s priced, and how it’s tracked. The challenge for enterprises is building that same level of confidence across hundreds of systems and millions of records.

“If the pitch that we need better data quality worked, then I wouldn’t be on your show, because people would be doing it. It would have been done. It wouldn’t be something that we’re still talking about.”

— Scott Taylor, Founder, MetaMeta Consulting

Why data leaders lose the room (and how storytelling wins it back)

I spent the first half of my career as a data practitioner and the second half in product marketing, so I’ve seen this from both sides. Data professionals are trained in hard skills, and nobody starts their data career with a course on how to pitch metadata management to a CFO. Marketing and sales teams learn to tell stories because their results depend on it, but data leaders tend to explain the mechanics first, walking into a meeting to describe the technical approach and losing their audience before they ever get to the business case.

“Data people love to get technical. They love to explain how it’s going to get done. And you just lose the business folks right away. A CEO, if you want money from them, they don’t care how it’s done until they understand why it’s important to the organization.”

— Scott Taylor, Founder, MetaMeta Consulting

Scott advises flipping the sequence. Every business leader has stated goals around growth, market expansion, customer experience, or operational efficiency. The data opportunities are already embedded in those objectives. If you listen to what your business leaders say they want to accomplish, you will find where the discipline fits. Their conversations focus on customers, brands, and markets, and all of those have a data element to them.

The 3V framework: Vocabulary, Voice, and Vision

Scott structures his advice around a framework he calls the 3V of data storytelling for data management. Most data practitioners already know the 3V of big data, and the parallel is deliberate.

Vocabulary. Get the words right. The language you use to describe data management to a CFO should be different from the language you use with your data engineering team. Terms like “master data management” and “reference data governance” mean nothing to someone whose primary concern is revenue growth or margin improvement. Scott recommends using words like strength, structure, and foundation rather than quality, because quality can feel subjective and emotional, almost like a complaint when what you need is a business case.

Voice. Everyone involved in making the case for data management needs to tell a consistent story. If the data team, the IT team, and the business analysts are all framing the problem differently, the message gets diluted before it reaches the people who control the budget. Scott calls this harmonizing to a common voice across the organization.

Vision. Connect every data activity to the strategic intentions of the enterprise. The pitch becomes impossible to ignore when you frame data investment as the enabler of business outcomes the leadership team has already committed to achieving.

“You’ve got to connect those dots between why we need metadata management in the context layer to the CEO’s initiative of expanding to new markets and becoming better partners with our customers.”

— Scott Taylor, Founder, MetaMeta Consulting

AI is not the Ozempic for data governance

At the Gartner D&A Summit in Orlando, Scott and I both noticed the same thing on the show floor. The vendor messaging was, in Scott’s words, “horrifyingly consistent.” Nearly every booth was leading with agentic AI, AI-native architecture, and context layers. Scott’s tongue-in-cheek response was classic Scott. As people posted about vendor after vendor emphasizing “context,” he started commenting on their posts with the same line: “context is the new oil.”

This hype cycle mania repeats every few years, from data mesh and data fabric three years ago to big data before that, and the cycle always ends with organizations realizing the new thing doesn’t work without solid data management underneath it.^[4] Scott’s colleague Malcolm Hawker coined a name for it: the “semantic pedantic cycle.”

The way Scott sees it, the belief that AI will solve the data management problem on its own is the latest version of this thinking. He calls this “AI is the Ozempic for data governance,” a line that got plenty of laughs at Gartner and in our bonus Data Puppets segment. AI can assist with certain data management tasks, and the organizational discipline of establishing truth in your data before deriving meaning from it still requires human leadership and commitment.

Data Puppets: using satire to say what executives need to hear

Think of it as Dilbert for the data world, except with puppets. The cast includes the CDO (Chief Dog Officer), his sidekick ITB who speaks exclusively in buzzwords, and a “Cat Sultant” from Meow-kinsey whose primary initiative is to generate more billing. Just as Scott Adams captured the absurdity of corporate life in ways that employees pinned to their cubicle walls, Scott Taylor captures the absurdity of enterprise data management in ways that data teams share in their Slack channels.

What started as a collection of data jokes has turned into a communication tool that Scott didn’t anticipate. People use the episodes in internal presentations to illustrate how the data team sounds to the business side. A Chief Dog Officer can say things that would come off as harsh from a human consultant, and people laugh first and then recognize the pattern in their own organization.

In the bonus Data Puppets segment at the end of our recording, Scott introduced A-Eye, a puppet character who attended the Gartner D&A Summit and had opinions about everything. When asked about data quality, A-Eye’s response captured an attitude that data leaders encounter constantly: “They’ve been whining about data quality ever since there was data. If it was that important, it would have been solved by now.”

“The number one reaction I got was, ‘this is just like my organization.’ People were really taking it seriously. They were like, ‘I showed this to my team to show this is how we sound to the business side.’”

— Scott Taylor, Founder, MetaMeta Consulting

Next steps

Scott’s approach offers a practical starting point for data leaders who are struggling to get executive support for foundational data work. The investment is in changing the conversation, which costs nothing beyond the willingness to rethink how you communicate.

- Craft your one-sentence pitch. Distill why data management matters to your organization into the fewest possible words. Frame it as a business statement that connects to what leadership has already said they want to accomplish, because a technical explanation won’t land. If you can’t say it in one sentence, you haven’t refined it enough.

- Audit your storytelling sequence. Are you leading with how (the technical approach) or why (the business impact)? If your presentations start with architecture diagrams and technology stacks, consider flipping the order. Open with the business objective, show how data enables it, and save the technical details for the appendix.

- Apply the 3V framework. Review the vocabulary you’re using with executive stakeholders. Swap subjective terms like “data quality” for structural language like “data foundation” and “data trust.” Align your team to a common voice so the message doesn’t fragment across departments. Make sure every data initiative you propose connects to the organization’s stated strategic vision.

Listen to the full conversation with Scott Taylor on the Data Faces Podcast.

Based on insights from Scott Taylor, Founder at MetaMeta Consulting, featured on the Data Faces Podcast.

Share TinyTechGuides

Leave a comment

Frequently asked questions

What does “truth before meaning” mean in data management?

Truth before meaning is the principle that organizations must establish trustworthy, well-governed foundational data before attempting to derive business insights from it. The “truth” layer includes master data, reference data, metadata, and data governance. The “meaning” layer includes BI, analytics, AI, and everything that interprets data for business decisions. Scott Taylor argues that skipping the truth layer is the primary reason AI and analytics initiatives underperform. Gartner estimates that poor data quality costs the average enterprise $12.9 to $15 million per year.

What is the 3V framework for data storytelling?

The 3V framework is Scott Taylor’s approach to helping data leaders communicate with executive stakeholders. It stands for Vocabulary (choosing business language over technical jargon), Voice (aligning the entire data organization around a consistent narrative), and Vision (connecting every data initiative to the company’s strategic objectives). The framework is designed to shift data management conversations from technical how-to discussions to business-impact narratives that resonate with CFOs and CEOs.

Why do data leaders struggle to get executive buy-in for data management?

Data leaders are trained in hard skills and tend to lead presentations with technical approaches rather than business outcomes. Executives in marketing, sales, and the C-suite are better storytellers by training and practice. When data leaders explain master data management architecture before explaining how it connects to revenue growth or market expansion, they lose their audience. Scott Taylor recommends flipping the sequence and leading with why the work matters to the business before explaining how it gets done.

Can AI fix data quality problems on its own?

No. Scott Taylor calls the belief that AI can solve data management problems without organizational discipline “AI is the Ozempic for data governance.” While AI can assist with specific data management tasks, it cannot replace the foundational work of establishing trusted master data, standard hierarchies, and consistent taxonomies. Research from IBM and MIT Sloan Management Review suggests companies lose 15 to 25 percent of revenue from poor data quality, and that cost scales as AI investments grow without addressing the underlying data.

What are Data Puppets, and why do they matter for data management?

Data Puppets is a satirical puppet series created by Scott Taylor that uses humor to expose common enterprise data dysfunction. Characters include the CDO (Chief Dog Officer), a buzzword-fluent sidekick named ITB, and a consultant from “Meow-kinsey.” The series works as a communication tool because satire creates a layer of separation that lets audiences absorb uncomfortable truths about their own organizations. People use the episodes in internal presentations to illustrate how data teams sound to business stakeholders.

Podcast highlights

[0:06] Scott’s background as the Data Whisperer and 30 years in the data space

[3:59] Truth before meaning: Scott’s entire data philosophy in three words

[6:04] Why data truth isn’t philosophical, and the supermarket scanner example

[7:56] The importance of storytelling and why data practitioners aren’t trained in it

[10:27] Has AI changed the conversation about data management, or is it the same cycle?

[13:08] How vendors performed at the Gartner D&A Summit in Orlando

[16:27] “Context is the new oil” and the semantic pedantic cycle

[19:54] Crafting a one-sentence data management story for a skeptical CFO

[22:59] The 3V framework: Vocabulary, Voice, and Vision

[25:37] Data Puppets: how satire reveals organizational dysfunction

[31:48] Why humor helps executives hear truths they’d otherwise dismiss

[34:24] Where to find Scott Taylor and the Data Puppets

Bonus: Data Puppets segment — A-Eye attends the Gartner D&A Summit

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

- Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

- Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

- The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

- The CIO’s Guide to Adopting Generative AI: Five Keys to Success

- Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

- The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Gartner. “How to Improve Your Data Quality.” Gartner, 2021. https://www.gartner.com/smarterwithgartner/how-to-improve-your-data-quality.

^[2]David Sweenor. “AI in 2025: Why 90% of Gen AI Projects Will Fail.” TinyTechGuides, March 22, 2025.

https://insights.tinytechguides.com/p/ai-in-2025-why-90-of-gen-ai-projects

^[3]IBM. “The True Cost of Poor Data Quality.” IBM Think, 2024. https://www.ibm.com/think/insights/cost-of-poor-data-quality.

^[4]David Sweenor. “AI in 2025: Why 90% of Gen AI Projects Will Fail.” TinyTechGuides, March 22, 2025.

https://insights.tinytechguides.com/p/ai-in-2025-why-90-of-gen-ai-projects

.

This prompt is not part of The PMM’s Prompt Playbook which has 30 ready-to-use prompts. If you’re looking for more cut-and-paste prompts, join the Substack! Paid subscribers receive new cut-and-paste prompts every week.

Get the PMM’s Prompt Playbook and Modern B2B Marketing today!

Need help with product marketing or prompts? Let me know.

Workflow Name: Consensus Path Planning and Deal Acceleration

Created by insights.tinytechguides.com

What this workflow does

I’ve watched deals die where every single stakeholder privately supported the purchase. No one objected. No one said no. The deal just stalled because nobody created a mechanism to turn all that private support into a collective decision.

When a deal goes quiet, sales teams usually blame a lack of motivation. The champion went dark. The economic buyer stopped responding. The technical team wants more data. These look like engagement problems. They’re structural failures. The committee never reached a specific consensus milestone, and nobody noticed because the pipeline report tracks stages, not alignment.

This workflow maps the path from first engagement to signed deal. It identifies the sequence of stakeholder engagements, the milestones the committee must reach, the mechanisms that build collective commitment, and the recovery protocols for when deals stall. The output is a deal architecture document that gives sales a stage-by-stage engagement plan and gives marketing visibility into where their content supports committee alignment.

This workflow does not create messaging, map the committee, or surface hidden objections. That’s what the first three workflows in this series do.

It answers one question: what is the specific path from fragmented interest to collective commitment, and what do you do when that path breaks?

Workflow steps summary

Step 0: Define inputs

Step 1: Define the consensus milestones

Step 2: Map the stakeholder engagement sequence

Step 3: Identify consensus-building mechanisms

Step 4: Design the stall recovery playbook

Step 5: Build the deal architecture timeline

Step 6: Assemble the consensus path plan

This is the fourth and final workflow in the buying committee series. It builds on the outputs from the first three:

• Buying Committee Decision Mapping Workflow

• Hidden Objection and Risk Surface Analysis

• Committee-Aware Messaging and Content Mapping

The workflow was created by insights.tinytechguides.com and connects to these existing workflows:

• Prompt Workflow: Voice of Customer

• Prompt Workflow: Competitive Landscape Mapping

• QuickStart Battlecard for Competitive Sales Wins

• Strategic Battlecard Workflow for Competitive Wins

Step 0: Define inputs

Before running the workflow, gather these inputs:

• {account_type} = target account profile or segment

• {deal_examples} = recent won, lost, and stalled deals

• {sales_inputs} = call notes, objections, deal commentary

• {customer_inputs} = outputs from Voice of Customer workflow

• {competitive_context} = outputs from Competitive Landscape Mapping

• {committee_map} = output from Buying Committee Decision Mapping Workflow

• {risk_analysis} = output from Hidden Objection and Risk Surface Analysis Workflow

• {messaging_map} = output from Committee-Aware Messaging and Content Mapping Workflow

• {avg_sales_cycle} = average sales cycle length for this deal size

• {sales_process} = current sales stages or methodology (e.g., MEDDPICC, Sandler, custom stages)

Step 1: Define the consensus milestones

A signed deal is the final milestone, but it is the result of many smaller consensus events that happen inside the buying organization. Before a committee can approve a purchase, they must collectively agree on a series of intermediate decisions. That the problem is worth solving. That the category of solution is right. That this vendor is the best fit. That the timing is right. That the terms are acceptable. Missing any of these intermediate consensus points creates a stall, and most sales teams cannot tell you which milestone their deal has actually reached versus which one they have assumed.

Prompt:

Role

You are a senior B2B sales strategist who designs deal architectures for complex enterprise sales involving multi-stakeholder buying committees.

Context

A signed deal requires the committee to reach consensus on five intermediate decisions: problem significance, solution category, vendor preference, timing, and terms. Missing any of these creates a stall that shows up as a “stuck” deal in the pipeline.

Task

Using {committee_map}, {risk_analysis}, {sales_inputs}, and {deal_examples}, define the internal consensus milestones for a purchase in {account_type}.

Format

For each of the five milestones (problem consensus, category consensus, vendor consensus, timing consensus, terms consensus):

• Description of what agreement looks like at this stage

• Key stakeholders who must be aligned

• Evidence they need to reach alignment

• The gatekeeper for this milestone

• The most common reason this milestone stalls

• The event or evidence that unlocks it

Deliver as a table. After the table, provide a narrative summary (3-5 sentences) describing the typical sequence and where the buying process most often breaks down.

Tone

Strategic and grounded in real-world sales dynamics. Write as someone who knows the difference between a milestone that has been genuinely reached and one that has been assumed.

Your pipeline isn’t stuck. It just skipped a milestone nobody noticed.

Step 2: Map the stakeholder engagement sequence

The order in which you engage stakeholders matters as much as the message. Engaging the economic buyer too early, before the technical team has validated feasibility, undermines credibility. Engaging procurement too late, after expectations are set without their input, creates adversarial dynamics. The right sequence builds momentum by creating a chain of internal endorsements where each stakeholder’s support makes the next engagement easier.

Prompt:

Role

You are a senior enterprise sales architect who designs multi-threaded engagement strategies for complex B2B deals with large buying committees.

Context

Buying committees do not engage in a linear sequence. Some stakeholders can be engaged in parallel. Others must be sequential. The wrong order creates friction that compounds across the deal cycle.

Task

Using the consensus milestones from Step 1, {committee_map}, and {sales_inputs}, design the optimal stakeholder engagement sequence for {account_type}.

Format

1- Engagement timeline table with columns: Stakeholder, First Engagement Timing, Engaged By, Goal of First Engagement, Prerequisites

2- Multi-threading strategy describing which stakeholders can be engaged simultaneously, which must be sequential, and where the champion should make introductions versus where sales should request meetings directly

3- Escalation triggers table with columns: Stall Signal, Escalation Action, Who to Involve, Anti-Pattern to Avoid

4- A text-based engagement sequence showing the flow across deal stages

Tone

Tactical and specific. Build this for a sales team executing on real deals, not a training manual.

Step 3: Identify consensus-building mechanisms

Individual stakeholder buy-in does not automatically become committee consensus. A committee where every member privately supports the purchase can still fail to act because no one created the mechanism for converting private support into collective commitment. Consensus requires structure. Shared evaluation criteria, visible commitment signals, and events where alignment is tested and confirmed rather than assumed.

Prompt:

Role

You are a senior organizational psychologist and B2B sales strategist who specializes in the dynamics of group decision-making within enterprise buying committees.

Context

The gap between individual interest and collective commitment is where most deals die. Consensus requires shared frameworks, commitment events, and coalition management, not just good meetings with individual stakeholders.

Task

Using the outputs from Steps 1-2, {committee_map}, and {risk_analysis}, identify the mechanisms that build consensus within this buying committee for {account_type}.

Format

1- Shared evaluation framework with 5-8 weighted criteria the committee can use to evaluate together rather than in isolation

2- Commitment escalation events table with columns: Event, Purpose, Success Signal, Facilitation Notes

3- Coalition-building tactics as a numbered list, each with a 2-3 sentence description covering alliance strengthening, friction neutralization, swing vote conversion, and blocker isolation

4- Commitment artifacts table with columns: Artifact, Associated Milestone, Owner, Why It Matters

Tone

Practical and psychologically informed. Write as someone who understands both the organizational dynamics and the tactical moves that convert understanding into action.

Step 4: Design the stall recovery playbook

Every complex B2B deal stalls at some point. The question is not whether stalls happen but whether the team can diagnose the cause and execute a recovery before the deal slips to “no decision.” Most sales teams treat stalls as a motivation problem when they are actually a structural problem. A specific consensus milestone was not reached, or a specific stakeholder’s concern was not addressed, and no one noticed until the deal went quiet.

Read more

I spend most of my time consulting with organizations that are trying to figure out what to do with AI, and teaching is a big part of that work. I’ve also conducted more than 35 interviews on the Data Faces Podcast with data leaders, practitioners, and technology executives. The question I hear in every engagement and nearly every episode is the same: how do I know the output is right? When a chatbot gives a questionable answer, someone catches it and moves on. An autonomous agent, on the other hand, might already be three decisions downstream before anyone notices the answer was wrong.

At the Gartner Data & Analytics Summit in Orlando, I sat down with Conor Jensen for an on-location episode of the Data Faces Podcast. Conor is the Global Field CDO at Dataiku, a data science and machine learning platform used by enterprise organizations to build, deploy, and manage AI projects. It’s a role shaped by an unusual path. He purchased Dataiku as a customer about ten years ago, spent seven years on the other side of the table, and now helps organizations avoid the mistakes he already made. He’d just come off Dataiku’s biggest product launch in the company’s 13-year history, and one observation from our conversation captured exactly what I’ve been hearing from clients.

“A far more dangerous thing than an agent that breaks is an agent that’s still functioning and giving the wrong answers.” — Conor Jensen, Global Field CDO, Dataiku

According to Gartner, only 6% of organizations have AI agents in production today, while 53% are still in exploration mode.^[1] The organizations racing to build agents have largely skipped the question of whether the ones they already have are performing.

About Conor Jensen

- Conor Jensen is the Global Field CDO at Dataiku. He purchased Dataiku as a customer about ten years ago, joined the company seven years later, and now helps organizations develop AI strategy and operational plans to get the most out of the platform. Before Dataiku, he worked as a data scientist and analytics leader.

- Key topics discussed: Dataiku CoBuild and vibe coding for data pipelines, Reasoning Systems for multi-step autonomous decisions, the Agent Management Platform for cross-platform observability, getting AI from pilot to production, and why perfect data is never coming

Everyone’s building agents, nobody’s solved production

I asked Conor why so many AI projects stall between prototype and production. He didn’t point to a single bottleneck. He described a pile of them that keeps growing.

“We haven’t solved any of that yet as an industry. We just keep putting more in the backpack.” — Conor Jensen, Global Field CDO, Dataiku

MLOps was supposed to get machine learning models into production. Then came LLMOps for large language models. Now the industry is talking about AgentOps. Each layer adds new complexity without resolving the one that came before it. Conor sees three barriers that keep organizations stuck. Deployment architecture is one, where something that works on a laptop or in a dev environment falls apart on the way to production. Organizational dynamics are another, including governance, trust, and change management, which he considers harder than any technical challenge. And then there’s data readiness, where teams wait for perfect data that will never arrive.

Gartner’s research reinforces how high the stakes are. Rita Sallam estimates that 70% of agentic AI use cases will fail to deliver expected value due to underinvestment in necessary foundations.^[2] Data availability and quality remain the number one barrier to AI implementation, cited by 30% of data management leaders.^[3] Gartner analyst Sarah Turkaly reinforced the point at the summit: “Data governance will be the single point of failure for organizations’ AI ambitions.”^[4]

Dataiku’s biggest launch targets every layer of the problem

The opening keynote from Adam Ronthal and Georgia O’Callaghan set the tone for the summit by framing AI value around three returns: return on intelligence, return on integration, and return on individuals.^[5] Dataiku positioned its announcement around that same framework, branding the launch as “The Platform for AI Success.” Conor walked me through three new products that Dataiku announced at the summit, each one targeting a different layer of the production problem.

Dataiku CoBuild brings vibe coding into the data platform, but the comparison to building a web app breaks down quickly. With a web app, you click a button and the page loads or it doesn’t. With a data pipeline, you get summary statistics and a model, but verifying the answer requires a level of inspection that 2,000 lines of generated Python won’t give you. CoBuild takes that generated code and renders it as visual workflows you can step through, edit, and validate. Conor, a data scientist himself, was candid about why this matters.

“Out of 2,000 lines of Python and a machine learning project, there’s probably like 40 that are what’s really, really important. The rest of it is, yeah, okay, did you pull the right data?” — Conor Jensen, Global Field CDO, Dataiku

CoBuild abstracts the boilerplate so you can focus on the 40 lines that determine whether the output is trustworthy. It launches in June 2026.

Reasoning Systems tackle a different gap. Conor used the example of a supply chain analyst who today pulls data from five different systems, consults with other teams, and makes a judgment call. Reasoning Systems layer process flows and context on top of data sources, then give an agent the ability to walk through the entire sequence. The key difference from RPA is that not every step is deterministic. Some require the agent to self-correct or stop entirely. Dataiku is building these for targeted use cases in specific industries rather than trying to solve everything at once.

The product Conor said he’s personally most excited about is the Agent Management Platform. Fifty-four percent of organizations are exploring or deploying goal-driven AI agents, according to Gartner.^[6] The question most CIOs should be asking is straightforward: how many agents do I have in production across all of my systems? With agents being built and deployed on Databricks, Salesforce, and dozens of other platforms alongside Dataiku, that question is hard to answer today.

“How do I manage all of my agents across my infrastructure, wherever they’ve been deployed? How do I make sure I know that they’re performing, not just functioning, but performing?” — Conor Jensen, Global Field CDO, Dataiku

Monitoring whether an agent is running is table stakes. You can do that with an API bus. The Agent Management Platform goes further by adding performance management, a semantic layer, and contextual understanding across every environment where agents are deployed. It evaluates whether agents are delivering the right business results across eight, ten, or twenty different systems. It goes GA in September 2026 and does not require being a Dataiku customer.

Conor had practical advice for organizations that feel stuck waiting for perfect data or an industry standard to emerge.

“News flash. There’s no such thing as perfect data, never will be. You have to just get moving.” — Conor Jensen, Global Field CDO, Dataiku

Only 12% of D&A leaders say they are fully prepared to carry out their mandate, according to Gartner’s 2026 CDAO survey.^[7] Conor’s point is that treating full readiness as a prerequisite for action is its own form of failure.

Production is the starting line

Conor Jensen has seen the Dataiku platform from both sides over the past decade, and that practitioner-turned-vendor perspective came through in every answer he gave. The industry has spent years talking about getting AI to production. The conversation at Gartner this year made clear that production is only the starting line. The harder work is knowing what happens after you deploy, and most organizations have no way to answer that question across their agent portfolio today.

The next time someone on your team proposes building a new agent, ask a different question first. Do you know how the ones you already have are performing?

Listen to the full conversation with Conor Jensen on the Data Faces Podcast.

Based on insights from Conor Jensen, Global Field CDO at Dataiku, featured on the Data Faces Podcast.

Share

Share TinyTechGuides

Leave a comment

Frequently asked questions

What is Dataiku’s Agent Management Platform? Dataiku’s Agent Management Platform provides cross-platform observability and performance management for AI agents deployed across any system, including those built outside of Dataiku. It goes beyond uptime monitoring by evaluating whether agents are delivering correct business results. The platform adds a semantic layer and contextual understanding so organizations can assess agent performance across eight, ten, or twenty different environments from a single view. It is scheduled for general availability in September 2026.

How is vibe coding a data pipeline different from vibe coding a web app? With a web app, you can visually confirm whether it works by clicking a button and seeing the result. With a data pipeline, AI generates thousands of lines of code that produce summary statistics and a model, but there is no simple way to verify the answer is correct. Dataiku CoBuild addresses this by rendering generated code as visual workflows that users can step through, edit, and validate rather than reading through 2,000 lines of Python.

What are Dataiku Reasoning Systems? Reasoning Systems layer process flows and business context on top of data sources to enable multi-step autonomous decisions. Unlike RPA, where every step is deterministic, Reasoning Systems allow agents to self-correct or stop when results fall outside expected parameters. Dataiku is building these for targeted use cases in specific industries, starting with manufacturing operations, with supply chain and financial risk scheduled for later in 2026.

Why do most agentic AI use cases fail? Gartner estimates that 70% of agentic AI use cases will fail to deliver expected value due to underinvestment in necessary foundations. The top barrier to AI implementation is data availability and quality, cited by 30% of data management leaders. Organizations also struggle with deployment architecture, governance, and change management. Gartner analyst Sarah Turkaly warned that data governance will be the single point of failure for organizations’ AI ambitions.

How many organizations have AI agents in production? According to Gartner research from January 2025 surveying 3,412 respondents, only 6% of organizations have AI agents in production. Fifty-three percent are still in exploration mode, and 25% are piloting. Fifty-four percent of organizations are exploring or deploying goal-driven AI agents, but most cannot answer how many agents they have running across their infrastructure or whether those agents are delivering correct results.

Podcast highlights

[0:00] Introduction at the Gartner D&A Summit and Dataiku overview

[1:27] Three new product announcements: CoBuild, Reasoning Systems, Agent Management Platform

[2:49] Dataiku’s evolution in the age of Gen AI

[3:30] Why AI projects stay stuck in pilot purgatory

[5:30] Deployment architecture that works from dev to production

[7:00] CoBuild, vibe coding, and why data pipelines are different from web apps

[8:26] Why even data scientists need better coding practices

[9:32] Reasoning Systems and autonomous multi-step decisions

[11:01] Agent Management Platform and cross-platform observability

[13:00] Monitoring vs. performance management for agents

[15:00] Opening the gates with governance and guardrails

[17:00] GA timeline, availability, and closing

About David Sweenor

David Sweenor is an AI advisor, author, and the founder of TinyTechGuides. He spent the first half of his career as a practitioner at IBM, building data warehouses and running predictive models, and the second half in product marketing leadership at SAS, Dell, TIBCO, Alteryx, and Alation. He advises Fortune 500 companies on AI strategy, data governance, and go-to-market planning, and hosts the Data Faces Podcast, where he interviews the leaders, practitioners, and technologists shaping the future of data and AI.

Books

- Artificial Intelligence

- Generative AI Business Applications

- The Generative AI Practitioner’s Guide

- The CIO’s Guide to Adopting Generative AI

- Modern B2B Marketing

- The PMM’s Prompt Playbook

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Chandrasekaran, Arun. “Navigating the AI Agent Landscape: A Strategic Guide for IT Leaders.” Gartner D&A Summit 2026, March 2026.

^[2]Sallam, Rita. “How to Calculate the Value and Cost of AI Agents.” Gartner D&A Summit 2026, March 2026.

^[3]Ramakrishnan, Ramke. “How Is Agentic AI Impacting and Disrupting Your Data Management Discipline?” Gartner D&A Summit 2026, March 2026.

^[4]Turkaly, Sarah. “The Future of D&A Governance.” Gartner D&A Summit 2026, March 2026.

^[5]Ronthal, Adam and Georgia O’Callaghan. “Navigate AI on Your Data & Analytics Journey to Value.” Gartner D&A Summit 2026 Opening Keynote, March 9, 2026.

^[6]Ramakrishnan, Ramke. “How Is Agentic AI Impacting and Disrupting Your Data Management Discipline?” Gartner D&A Summit 2026, March 2026.

^[7]Gabbard, Michael. “Signature Series: State of D&A 2026.” Gartner D&A Summit 2026, March 2026.

The spreadsheet might still be the most widely used data catalog on the market. That’s not a joke. It’s a finding from Stewart Bond, Research VP at IDC, who has spent the past decade studying how companies manage intelligence about their data. When I sat down with Stewart on the Data Faces podcast, he pointed out that every survey he runs surfaces the same contradiction. Organizations rank data quality as their top AI concern, yet they fail to invest in the one technology category designed to address it.

The frustrating part is that the data teams usually know exactly what the problem is. They flag quality and governance issues, but the budget continues to flow towards AI model development and agents instead. Stewart has been tracking this gap longer than most, and his perspective on how data intelligence evolved from an analyst’s shorthand into a global market category offers a useful lens for understanding why the gap persists.

“One of the biggest challenges organizations have is managing the intelligence about their data. Data catalogs, business glossaries, data lineage, all that stuff is so important now as we get into AI. And yet, their top investment categories are not on data catalogs.” — Stewart Bond, Research VP, IDC

About Stewart Bond

Stewart Bond is a Research VP at IDC, where he leads the data intelligence and data integration software research practice. His career spans over 30 years in IT, including a decade as a certified IT architect at IBM before moving into industry analysis in 2011. Outside of work, Stewart is a competitive curler who came within one match of representing Ontario at a Canadian national championship. In our conversation on the Data Faces Podcast, we discuss:

• How Stewart coined the term “data intelligence” and watched it become a global market category

• The difference between intelligence about data and intelligence from data

• Why agentic AI demands a shift-left approach to data quality

• What CDOs are most concerned about and where they’re under-investing

How one research note became a market category

Stewart joined IDC in 2015 and inherited a research area covering data integration and data access, which at the time included eight sub-markets like metadata management, data quality, and master data. A conversation with ASG Technologies introduced him to their term “enterprise data intelligence.” Stewart saw something useful in the phrase but dropped the “enterprise” qualifier. Data intelligence, as a simpler label, stuck.

The real momentum came in 2018, when GDPR was about to take effect. Enterprise data leaders started calling Stewart with the same question. “Where can I buy a data governance solution?” His response surprised them. You can’t buy governance. Data governance is an organizational discipline that requires people, processes, and accountability. What you can buy is data intelligence technology, the tools that tell you everything you need to know about your data so you can govern it.

“I had a lot of end-user clients calling me and saying, ‘Where can I buy a data governance solution?’ And I just kind of laughed, because data governance isn’t a technology solution.” — Stewart Bond, Research VP, IDC

Stewart framed this through his 5 W’s of data. Who is using it? How is it being used? Where does it live? What does it mean? Why do you even have it? How long do you have to keep it? These questions form the foundation of effective data governance, and answering them requires technology that most organizations still haven’t fully invested in.^[1]

Intelligence about data vs. intelligence from data

The term spread faster than Stewart expected. Collibra became “the data intelligence company.” Erwin (now Quest) adopted it for their data catalog. Alation started using it in 2020, after learning the phrase wasn’t a Collibra trademark but an industry-level concept. Informatica wove it into their intelligent data platform messaging. Then, in late 2023, Databricks made a major push with its own version of data intelligence.

The Databricks definition, however, expanded the original meaning. Stewart had always treated data intelligence as intelligence about data. What is this data, where did it come from, who uses it, and how good is it? Databricks extended the concept to include intelligence from data, using the metadata and context layer to generate smarter analytics and AI outcomes from the data itself. The distinction matters because it changes what organizations expect from the category and how they evaluate platforms.

Dave Kellogg was serving as acting CMO at Alation when he first explored the term’s origins with Stewart. After the Databricks announcement, Kellogg reached out with a direct assessment. “I think you did it. I think you created a new market category.” Last year, IBM confirmed the trend by rolling its entire portfolio of data cataloging, quality, lineage, and observability products into IBM watsonx Data Intelligence. IBM’s product leadership told Stewart the renaming was a direct result of his work and the broader market momentum he helped create.

“I’d always treated data intelligence as intelligence about the data. I’d say Databricks has extended it to intelligence from the data, getting more into the case of leveraging that intelligence about the data to make sure you’re using the data intelligently.” — Stewart Bond, Research VP, IDC

Agents can’t wait for clean data

The shift to agentic AI fundamentally changes how organizations need to approach data quality. Traditional analytics workflows gave organizations a buffer. Data moved through batch processes, giving teams time to spot anomalies and intervene before a bad number reached a dashboard. Autonomous agents don’t offer that luxury. An agent monitoring a change data capture stream sees a new order event and starts fulfilling it on the spot. If the data in that event is wrong, the agent acts on it before anyone has a chance to review it.

Stewart describes this as the “shift left” imperative. Data quality, privacy, and integrity all need to move as close to the source as possible, because once data enters the agentic pipeline, there is no batch window to clean it up. Deloitte flagged this as one of four critical data quality challenges for AI, finding that companies building agentic systems need quality controls embedded at the point of data creation, not applied after the fact.^[2]

“You’d better make sure the data in that order event is good and that it’s a real and reliable order event. You may have heard the term shift left. Your data quality, your data privacy, your data integrity all need to be as close to the source as possible.” — Stewart Bond, Research VP, IDC

The challenge extends beyond structured data. Stewart raised a question that most organizations still haven’t answered well. What do you do about the unstructured data that makes up the bulk of enterprise information? Every organization has countless versions of the same PowerPoint file, thousands of PDFs, and documents that LLMs are eager to ingest. Some vendors are starting to crack this problem. Shelf.io, for example, has developed methods to assess the quality of unstructured documents, a capability that seemed impossible just a few years ago.

The broader issue remains, though. Most organizations lack the data context needed to determine whether their unstructured data is safe to use, let alone high-quality.^[3] Stewart sees agentic AI as part of the eventual solution. Agents that pre-populate data catalogs and reduce the manual burden on data stewards could finally solve the adoption problem that has held these tools back for years. But that future depends on investing in the foundation today.

The investment gap CDOs can’t ignore

Stewart runs an annual survey of the Office of the Chief Data Officer, and the results tell a consistent story. When you ask CDOs what their biggest organizational concern is, skills top the list. They struggle to find people who can do the work. The second concern is managing expectations around what AI can deliver, not just within their own teams, but across the C-suite, where leadership is under pressure to show results quickly and often treats AI as a magic bullet.

“Their top investment categories are not on data catalogs. Back to the spreadsheet might still be the most widely used data catalog on the market. I don’t have data to prove that, but anecdotally, that could be the case.” — Stewart Bond, Research VP, IDC

What makes this frustrating is that CDOs now have more influence over IT spending than ever before. IDC predicted in 2024 that chief data officers would gain significantly more budget authority by 2025, driven by the fact that every major AI concern in enterprise surveys points to data: quality, correctness, privacy, and security. CDOs are accountable for all of it. Deloitte’s 2025 CDO Survey tells a similar story. These leaders are increasingly expected to demonstrate direct business impact from their data programs, even as their organizations resist the investments required to achieve it.^[4]

And yet, when Stewart looks at where enterprises are actually putting their money, the top investment categories are not data catalogs or data quality tools. The data lineage, metadata management, and business glossary capabilities that form the backbone of data intelligence remain underfunded, even as AI programs depend on them.^[5] That spreadsheet Stewart mentioned at the top of our conversation? For many organizations, it is still doing the job that a proper data catalog should be doing.

You’ll never score 100%

Stewart closed our conversation with an insight he picked up years before he even joined IDC. A life insurance company told him they had finally accepted that their data would never be 100% clean. Instead of chasing perfection, they started measuring how clean or dirty their data was and feeding that score into their calculations. Their actuaries knew how to work with uncertainty. They just needed the number.

Data intelligence doesn’t promise perfect data. It gives you a clear picture of how much you can trust what you have. Organizations that know the quality of their data before it enters an AI pipeline avoid the costly cycle of debugging outputs that were doomed from the start. A data quality score of 75 means something different from a score of 95, and both are more useful than no score at all. When that score travels alongside the data into an AI model or an autonomous agent, the organization can make informed decisions about how much confidence to place in the output.

Start with Stewart’s 5 W’s. Audit how your organization currently tracks who uses its data, where it lives, and how trustworthy it is. If the answer is a spreadsheet, you have your business case.

The spreadsheet is still winning. It doesn’t have to be.

Listen to the full conversation with Stewart Bond on the Data Faces Podcast.

Based on insights from Stewart Bond, Research VP at IDC, featured on the Data Faces Podcast.

Share

Leave a comment

Frequently asked questions

What is data intelligence?

Data intelligence is the category of technology that provides intelligence about your data. It encompasses data catalogs, business glossaries, data lineage, data quality, and metadata management. Stewart Bond, Research VP at IDC, coined the term to describe the tools that answer foundational questions about data, including who uses it, where it lives, what it means, and how trustworthy it is. More recently, vendors like Databricks have expanded the definition to also include intelligence from data, using that context layer to improve analytics and AI outcomes.

How does data intelligence differ from data governance?

Data governance is an organizational discipline that requires people, processes, and accountability. Data intelligence is the technology that supports it. You cannot buy a data governance solution, but you can invest in data intelligence tools that tell you everything you need to know about your data so you can govern it. Organizations that try to solve governance with technology alone tend to fail, according to IDC’s Stewart Bond.

Why does agentic AI require a shift-left approach to data quality?

Traditional analytics workflows gave teams time to spot and fix data issues in batch processes before results appeared on the dashboard. Autonomous AI agents operate in real time and act on data the moment they receive it, with no batch window to clean things up. This means data quality, privacy, and integrity controls need to move as close to the data source as possible. Deloitte identified this as one of four critical data quality challenges for organizations building agentic AI systems.

What are CDOs most concerned about in 2025?

According to IDC’s annual survey of the Office of the Chief Data Officer, skills gaps rank as the top concern. CDOs struggle to find qualified people to do the work. The second biggest concern is managing leadership expectations around what AI can realistically deliver. Despite growing influence over IT budgets, CDOs face a persistent disconnect between the data foundation AI requires and where their organizations actually invest.

Where are organizations under-investing in data intelligence?

IDC survey data show that the top enterprise investment categories are not data catalogs, data quality tools, or data lineage capabilities, even though managing data intelligence is one of the biggest challenges organizations report. Stewart Bond notes that the spreadsheet may still be the most widely used data catalog on the market, a sign that foundational data intelligence technology remains significantly under-funded relative to AI program spending.

Podcast highlights

[0:05] Introduction and Stewart’s background at IDC

[2:31] Stewart’s life outside work, competitive curling, and fishing

[5:00] The origin of the term “data intelligence” and the ASG Technologies connection

[6:44] GDPR drives demand for governance solutions, the 5 W’s of data

[8:15] Collibra, Erwin, Alation, and Informatica adopt the term

[10:00] Databricks expands the definition, Dave Kellogg’s “you created a category” moment

[14:00] IBM rebrands to watsonx Data Intelligence

[18:00] Intelligence about data vs. intelligence from data

[26:00] Agentic AI and the shift-left imperative for data quality

[29:00] Unstructured data quality and Shelf.io

[31:00] What CDOs are most concerned about in 2025

[35:00] Where organizations are under-investing in data intelligence

[36:40] Data quality will never be 100%, the life insurance anecdote

[38:00] Agentic AI and the future of data catalog adoption

About David Sweenor

David Sweenor is a Top 25 AI thought leader, six-time author, and founder of TinyTechGuides. He spent the first half of his career as a practitioner at IBM, building data warehouses and running predictive models, and the second half in product marketing leadership at SAS, Dell, Quest, TIBCO, Alteryx, and Alation. He hosts the Data Faces Podcast, where he talks with the people who are making data, analytics, and AI work in the real world.

Books

- Artificial Intelligence

- Generative AI Business Applications

- The Generative AI Practitioner’s Guide

- The CIO’s Guide to Adopting Generative AI

- Modern B2B Marketing

- The PMM’s Prompt Playbook

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.

^[1]Sweenor, David. “Why the Biggest AI Enthusiasts Care Most About Governance.” TinyTechGuides, January 27, 2026. https://tinytechguides.com/blog/why-the-biggest-ai-enthusiasts-care-most-about-governance/

^[2]Deloitte. “Four Data and Model Quality Challenges for AI.” Deloitte AI Institute, 2025. https://www.deloitte.com/global/en/our-thinking/insights/topics/artificial-intelligence/ai-data-quality-challenges.html

^[3]Sweenor, David. “Your AI Doesn’t Have a Model Problem. It Has a Data Context Problem.” TinyTechGuides, February 24, 2026. https://tinytechguides.com/blog/your-ai-doesnt-have-a-model-problem-it-has-a-data-context-problem/

^[4]Deloitte UK. “CDO Survey 2025.” Deloitte United Kingdom, 2025. https://www.deloitte.com/uk/en/services/consulting/analysis/chief-data-officer-survey.html

^[5]Sweenor, David. “Data Lineage for AI: Why Truth Beats Hope in Banking.” TinyTechGuides, December 2, 2025. https://tinytechguides.com/blog/data-lineage-for-ai-why-dotrth-beats-hope-in-banking/

I walked the expo floor at the Gartner Data & Analytics Summit in Orlando expecting every conversation to be about AI, agents, and context layers. They were, and plenty of vendors had agent-washed their messaging overnight. The vendors were talking about the future, but the practitioners were pointing to something they already had.

The opening keynote set the tone. Adam Ronthal and Georgia O’Callaghan reported that four out of five organizations are now deploying AI, but only one in five will achieve their stated ROI.^[1] Governance, they argued, is a value accelerator and should be treated as one. With AI agents on every vendor’s booth and in nearly every session title, the question of how to govern autonomous systems had real urgency behind it.

I carried that framing into on-location interviews for the Data Faces Podcast with Philip Dutton, CEO and founder of Solidatus, Terrence Hedin, Data and Metadata Platform Director at the London Stock Exchange Group, and Caleb Watkins, Solutions Engineer at Solidatus. Three different roles, three different vantage points, and they all pointed to the same thing. The most valuable AI governance asset many organizations have is the data lineage and metadata infrastructure that their compliance teams built years ago. Solidatus, a data lineage and metadata management platform used by financial services and other regulated industries, served as the common thread across all three conversations.

From second-class citizen to strategic asset

I suggested to Terrence Hedin that before AI changed the conversation, lineage and metadata were treated as second-class citizens. He expanded on that.

“It has evolved. It is a first-class citizen,” Terrence said. “Every business requirement spec includes lineage at an element level. Every tech spec includes how you produce that lineage.”

At LSEG, lineage used to answer a narrow set of questions. Where did this data come from? Can we prove it to regulators? Those questions still matter, but Terrence described how LSEG now brings business metadata, technical metadata, and semantic layers together into a knowledge graph that serves the entire organization.

“We bring our business metadata, our technical metadata, our semantic layers, into a knowledge graph so we can build that true business context. That provides not only human benefit, but machine benefit as well.” — Terrence Hedin, Data and Metadata Platform Director, LSEG

LSEG now treats metadata as a data product, published to both internal teams and external customers. Not a theoretical data mesh exercise, but a commercial product. The governance infrastructure they built for compliance became the foundation for a revenue-generating line of business.

Gartner research supports this trajectory. In the session “Trust as the New Currency,” Guido De Simoni presented data showing that organizations with graduated trust models achieve 64% compliance success compared to 23% without them.^[2] The trust frameworks that organizations like LSEG built for regulators directly support AI readiness.

Caleb Watkins, a Solutions Engineer at Solidatus, showed me a related capability. Because Solidatus centralizes all data and metadata in one place, organizations can load their regulations as reference models and let the AI assistant evaluate compliance across their data landscape.

“We can train Solidatus up on those regulations, and then we can ask the assistant to assess your models for compliance with these different regulations to make sure that you’re meeting all of your objectives.” — Caleb Watkins, Solutions Engineer, Solidatus

Lineage is no longer just a record of where data came from. It’s becoming the system that evaluates whether your data meets the obligations attached to it.

Your compliance operating model already works for AI

I said something to Philip Dutton that surprised us both. “Something that was built for compliance is now incredibly useful for AI.”

Philip didn’t hesitate. Whether it’s an AI consuming data, a BI dashboard pulling reports, or another system sharing information across business lines, the obligations are the same. Purpose limitations, storage rules, and sharing boundaries all travel with the data.

“You don’t have to change your operating model for AI governance. You can use the same operating model that you’ve been using, which the organization knows, and it takes them a long time to get to know it and to feel comfortable with it. So this really gives you a nice accelerator.” — Philip Dutton, CEO and Founder, Solidatus

The program already exists, and your teams know how to run it. The organizational trust has been earned over years of practice. Rather than standing up a parallel AI governance function, extend the operating model you already have.

Gartner analyst Andrés García-Rodeja reinforced this point in the session “How to Build the Context Layer for Reliable AI Agents.” By 2028, he estimates, 60% of agentic analytics projects relying solely on the Model Context Protocol will fail due to the lack of a consistent semantic layer.^[3] The metadata and lineage infrastructure that compliance teams maintain is exactly the kind of semantic foundation AI agents need to operate reliably. For AI teams building production pipelines and deploying agents, the implication is direct: the semantic layer your models need may already exist in your governance program.

And that operating model is getting faster. Caleb walked me through code scanning with the AI Lineage Assistant, which reduces what used to take several days of manual analysis to five to ten minutes, with 10x to 100x acceleration across broader governance workflows. In the session “Using Active Metadata to Support Data Agents,” Gartner analyst Mark Beyer presented research showing that metadata volume grows exponentially with agentic AI.^[4] Manual approaches to lineage and governance won’t survive that scale. Organizations like LSEG that automated their metadata workflows early have a compounding advantage over those still relying on spreadsheets and tribal knowledge.

AI trust starts with what you can see

Every conversation I had at the summit circled back to trust. De Simoni found that more than 50% of vendors identify trust as the top barrier to agentic AI adoption.^[5] Gartner expects unsupervised AI deployment to remain below 10% through 2028. The industry is building AI agents faster than it’s building the trust infrastructure to support them.

Philip put it simply: “If we can’t see it, if we can’t understand it, how do we trust it?” Solidatus renders data lineage as interactive visual maps rather than rows of metadata in a spreadsheet. Visualization isn’t a nice-to-have for governance. When people can see their data lineage mapped out and confirm it matches their understanding of the organization, they trust it. When they’re poring over raw metadata for hours, they generate questions, not confidence.

That principle extends to AI outputs as well. Solidatus built hallucination protection directly into the AI Lineage Assistant. If the LLM returns a response that isn’t grounded in metadata within the platform, the system rejects it and forces a new attempt. The response has to be anchored in real data before it reaches the user. In financial services and other regulated industries, where human-in-the-loop oversight is standard, that validation layer is non-negotiable.

Terrence described how trust and lineage connect at enterprise scale. LSEG’s data trust program is built on four elements of trust, with Solidatus providing the lineage foundation.

“If we don’t understand what that data is, it’s very difficult for us to understand how we can use it, how we should use it, what value it can provide,” Terrence said.

Trust becomes even more critical as AI agents grow more autonomous. Philip pointed out that much of what vendors call “AI agents” today are chatbots running on request-response. True agentic AI creates its own plan, executes across 20 to 50 steps, and self-corrects along the way. Without lineage and metadata infrastructure, organizations have no way to verify what an agent did or why.

What to do with the infrastructure you already have

The data lineage and metadata systems that compliance teams built over the past decade are becoming the critical infrastructure layer for AI trust, AI agents, and AI governance. LSEG proved that by turning their lineage program into a strategic asset and a commercial data product. Solidatus proved it by extending a governance platform into an AI-accelerated workflow engine.

If your organization has invested in data lineage for compliance, the next step isn’t building a separate AI governance program. Audit what you already have. Identify where it covers AI use cases. Close the gaps. If you lead an AI or data science team, ask your governance counterpart what lineage coverage already exists for your training data, production models, and agent workflows. The organizations that connect these functions now will govern AI with confidence. The ones that start from scratch will spend the next two years catching up.

Listen to the full conversations with Philip Dutton, Terrence Hedin, and Caleb Watkins on the Data Faces Podcast.

Based on insights from Philip Dutton, CEO and Founder at Solidatus, Terrence Hedin, Data and Metadata Platform Director at LSEG, and Caleb Watkins, Solutions Engineer at Solidatus, featured on the Data Faces Podcast.

Frequently asked questions

What is Solidatus?

Solidatus is a data lineage and metadata management platform that maps, visualizes, and governs data flows across the enterprise. It is used primarily by financial services and other regulated industries to track how data moves through systems, meet compliance obligations, and build organizational trust in data. The platform recently introduced an AI Lineage Assistant that adds natural language interaction, automated code scanning, and regulatory compliance assessment to its existing governance capabilities.

Why is data lineage important for AI governance?

Data lineage documents where data comes from, how it moves through systems, and what obligations are attached to it. Those obligations, including purpose limitations, storage rules, and sharing boundaries, apply to AI the same way they apply to BI dashboards or regulatory reports. Organizations with mature lineage programs can extend their existing governance operating model to cover AI use cases without building a separate framework. Gartner research presented at the 2026 D&A Summit showed that organizations with graduated trust models achieve 64% compliance success compared to 23% without them.

How does data lineage differ from AI governance?

Data lineage is a component of AI governance, not a separate discipline. Lineage tracks how data flows through an organization and what happens to it along the way. AI governance addresses the broader question of how to ensure AI systems use that data responsibly. The argument from practitioners at LSEG and Solidatus is that the lineage and metadata infrastructure built for regulatory compliance already provides the semantic foundation AI agents need. Rather than creating a parallel AI governance program, organizations can extend what they have.

What is a bring-your-own-LLM model for data governance?

A bring-your-own-LLM model allows organizations to connect their own large language model to a governance platform rather than sending data through a vendor’s AI infrastructure. Unlike vendor-hosted AI models that route customer data through external systems, the BYOLLM approach keeps all data processing within the customer’s own environment. Solidatus uses this approach for its AI Lineage Assistant, meaning no data flows through Solidatus or any third party. This design addresses the primary security concern enterprises have about AI in governance contexts, particularly in regulated industries like financial services.

How does Solidatus prevent AI hallucinations in governance workflows?

Solidatus built hallucination protection directly into the AI Lineage Assistant. When the LLM generates a response, the system validates it against metadata that exists within the platform. If the response isn’t grounded in real data, the system rejects it and forces a new attempt. The response has to be anchored in verified metadata before it reaches the user. This approach ensures that AI outputs in governance contexts are based on actual organizational data rather than fabricated information.

Where should organizations start with AI governance if they already have data lineage?

Start by auditing your existing lineage coverage to identify where it already applies to AI use cases. Philip Dutton, CEO of Solidatus, argues that organizations don’t need a new operating model for AI governance because the one they already use for compliance works. LSEG provides a proof point, having evolved their lineage program from a regulatory tool into a strategic asset and commercial data product. The key is closing gaps rather than starting from scratch.

Podcast highlights

Philip Dutton, CEO and Founder, Solidatus (~15 min)

[0:00] Introduction at the Gartner D&A Summit

[0:28] What is Solidatus and why data lineage matters

[0:54] Data lineage meets AI governance

[1:50] The AI Lineage Assistant and natural language interaction

[3:17] Trust in AI and trust through lineage

[4:45] Human in the loop for financial services

[5:14] Why visualization builds data trust

[7:05] You can’t automate what you don’t understand

[8:27] Data lineage as AI lineage, same operating model

[9:29] What’s on attendees’ minds at Gartner

[10:48] True agentic AI vs. chatbots

[12:00] The future of Solidatus and agentic orchestration

[14:18] LSEG session preview and closing

Terrence Hedin, Data and Metadata Platform Director, LSEG (~6 min)

[0:00] Introduction and upcoming LSEG session preview

[0:23] Overview of the LSEG talk with Philip Dutton

[2:26] Lineage as a first class citizen in the age of AI

[3:32] From regulatory reporting to strategic asset

[4:47] How the Solidatus AI Lineage Assistant is changing workflows

[5:54] Session details and closing

Caleb Watkins, Solutions Engineer, Solidatus (~4 min)

[0:00] Introduction at the Gartner D&A Summit

[0:22] The AI Lineage Assistant and bring-your-own-LLM

[0:44] Trust and security in the AI agent

[1:01] Use case: AI-powered code scanning

[1:42] Days to minutes with automated lineage

[2:01] Use case: regulatory compliance (BCBS 239, AI Act)

[2:56] Customer feedback on the assistant

[3:29] Find Solidatus at Booth #929

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

• Artificial Intelligence: An Executive Guide to Make AI Work for Your Business

• Generative AI Business Applications: An Executive Guide with Real-Life Examples and Case Studies

• The Generative AI Practitioner’s Guide: How to Apply LLM Patterns for Enterprise Applications

• The CIO’s Guide to Adopting Generative AI: Five Keys to Success

• Modern B2B Marketing: A Practitioner’s Guide to Marketing Excellence

• The PMM’s Prompt Playbook: Mastering Generative AI for B2B Marketing Success

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn

^[1] Ronthal, Adam, and Georgia O’Callaghan. “Opening Keynote: The State of Data and Analytics.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[2] De Simoni, Guido. “Trust as the New Currency.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[3] García-Rodeja, Andrés. “How to Build the Context Layer for Reliable AI Agents.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[4] Beyer, Mark. “Using Active Metadata to Support Data Agents.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.

^[5] De Simoni, Guido. “Trust as the New Currency.” Gartner Data & Analytics Summit, March 9-11, 2026, Orlando, FL.