
If you’ve sat through a vendor pitch meeting in the last six months, you’ve probably heard “context engineering,” “the semantic layer,” or “we need more context.” In plain terms, context engineering gives an AI system the facts, rules, and relationships it needs before answering. Every data platform and AI governance vendor has a slide on it now. The vocabulary is new, but the underlying problem is 30 years old.
I know because three people with deep data chops told me the same thing on the Data Faces Podcast this year, months apart, without talking to each other.[1]
Scott “The Data Whisperer” Taylor watched Gartner’s conference posts about context roll in and started commenting on them. His line: “context is the new oil.”
Steve Wooledge, CEO at Collate, put it this way: “Everybody talks about context now. Context engineering is the next new thing. It’s almost like data scientists back in the big data movement.”
Malcolm Hawker, chief data officer at Profisee, started with the generous read: “Context is great. You need it. In and of itself, LLMs cannot just consume raw tables of data.”
The new vocabulary gives the data community a common term for an old problem. It doesn’t give the marketing function anything at all.
What “context engineering” means, and why it has a name now
The term replaced “prompt engineering” in the buzzword bingo lexicon the analysts sell. Gartner’s July 2025 “First Takes” note on the topic was titled “Lead the Shift to Context Engineering as Prompt Engineering Fades,” and the summary says: “Context engineering is in, and prompt engineering is out.”[2] Prompt engineering dropped off Gartner’s 2025 Hype Cycle for Emerging Technologies.
Context engineering is the work of giving an AI system the facts, rules, and relationships it needs before it answers a question. Underneath the new parlance are older ideas.
A semantic layer is a shared definition of what a customer, a product, or a revenue number is. Master data is the one agreed-upon record for each real-world thing. The data world has spent three decades building these. AI is now the loudest reason to solve the problem, so you’ll hear about it in every pitch for the next year.
Context engineering hype isn’t targeted towards marketing
Every one of those pitches is going to the CDO’s budget, not yours.
The vendor roadmap for context engineering is a data-team roadmap. A catalog and a lineage tracker underneath, and then a semantic layer with metadata governance on top. The moves inside the category are big and getting bigger. Wooledge pointed to Salesforce acquiring Informatica as evidence that the big platforms want the metadata layer that connects everything, not just the integration plumbing underneath it. He’s right about where the market is moving.
The marketing function has a different context problem. The AI jobs you want running don’t need to know what “net revenue” means across your general ledger and your Looker dashboards. They need to know your positioning and your voice, which of three battlecards for a competitor is current, what your reps caught the last one getting wrong, and who sits on the buying committee at your top five accounts. None of those lives inside a semantic layer, and none of them are for sale.
I promised context engineering would get its own post in my AI GTM stack piece last month. This is that post. It’s not the one your vendor is writing.
The two halves of context, and why neither is yours
Hawker has a sharper way of saying what’s inside the term. On the same episode, he described what most vendors mean by context: “a semantic layer, which are the things that define kind of key terms… what a customer is, defines what net revenue is, defines what even a location is.” He called that the T-box, borrowing a knowledge-management term for terminology. Then he named the half that isn’t on any slide: “what a knowledge engineer would call the A-box, which is the actual values of customer, the values of product.”
In plain terms: the T-box tells your AI what a “customer” means. The A-box knows there are 15 David Sweenors in your CRM and which one is the current one. As Hawker put it, “If you’ve got 15 different versions of David Sweenor, all the context in the world isn’t going to solve that problem.”
A data team is trying to solve this two-sided problem for AI agents working on company data. It’s a real problem, but neither side of it is yours.
The marketing team’s context stack
The marketing function’s context stack is a list of files on disk. Not a platform, not a product, not something a vendor can hand you. A list.
Positioning doc: One page, in plain English. Who you are and who you are not, what you help with and what you refuse. The one-page cap matters because every AI job that writes something for you reads this file, and a longer file gets skimmed in patches. I have one for TinyTechGuides and a different one for every client I advise.
Voice guide: My voice guide runs 40 flagged phrases and 20 preferred constructions, built up over 10 months of my own edits. It tells your AI what you sound like: your banned phrases, your cadence rules, and the sentences that passed review last quarter next to the ones that got killed. When a cold agent reviews a draft with no context, it reverts to this file’s rules before anything else.
Field-correction log: Every claim your reps catch the AI getting wrong goes here. A rep says, “We don’t offer month-to-month contracts anymore; where did that come from?” and the correction goes in the log with a timestamp. The field-correction log outranks any published research, any analyst report, or any vendor datasheet. Three months in, this is the file your competitors don’t have, because they don’t have your reps.
Master-file index: A single file that names every other master file, one per fact. The master positioning doc lives at this path. The master competitor-pricing sheet lives at this path. Everything else is a draft or a copy. If three files mention Competitor X’s price, the index names which one is current and marks the others as archive or derivative.
Buyer-committee map: The file no one writes. Who approves a deal of this size, who objects, and what the champion needs to hear to sell it inside the company. This is the context your sellers hold in their heads for your top accounts. It’s also the context your AI jobs don’t have unless you write it down.
Scott Taylor’s three-word data philosophy sums up the point:
Truth before meaning. You’ve got to determine the truth in your data before you derive any meaning out of it.
— Scott Taylor, “The Data Whisperer,” Data Faces, Ep 35
You can’t give your AI meaning if it reads the wrong file first.
The master-file problem, scaled to your drive
In early September, I built a Claude skill to update my content calendar. It pulled from a Google Sheet, found the row I’d named, and rewrote two fields. The code worked. The sheet existed. The context (my calendar structure, my column rules, my overall file layout) was all written down.
It overwrote two rows and created a duplicate before I caught it.
The sheet had a legacy article-ID column that the skill read as if it were the row number, and the sheet wasn’t sorted by it. Nothing in my context said, “row IDs are in column A, not column B.” The skill guessed. The context existed. The rule that would have let the skill use the context didn’t.
Now picture the same mistake inside your marketing team’s Google Drive. A PMM asks Claude to rebuild the Competitor X battlecard. Three files mention Competitor X’s pricing. One is from 18 months ago, a snapshot from the last win-loss review. Another is a battlecard draft from a product marketer who left six months ago. The third is current, updated last Tuesday by the analyst relations lead after Competitor X’s earnings call. Nothing tells the agent which is which.
Without a master-file rule, it picks the one highest in the directory. Or the one with the earliest filename. Or the one someone most recently opened. Either way, your rep gets an out-of-date claim on a battlecard they’ll quote to a prospect. That’s the same identity problem Hawker named earlier, moved from a semantic layer into a shared drive. You have to name your master files yourself.
Four things to do this week
Name your master files. For every fact your AI jobs read (competitors, accounts, pricing, voice, and positioning), pick the single file that holds the current version. Mark every other copy as archive or derivative. If you can’t find one, make one, and put its path in a file called master-files.md at the top of your marketing drive.
Start the field-correction log. The next time a rep flags something an AI-written draft got wrong, add the correction to a log with a timestamp and a name. Do it for 30 days. The file is boring in week one, useful in week four, and the artifact your competitors can’t replicate by month three.
Write the one-page positioning doc. Not the brand book. Not the sales deck. One page, in your voice, that tells an AI job what you help with and what you refuse. If it runs longer than one page, your AI will read it in patches, and your drafts will drift.
Give the master-file index one owner. The context stack doesn’t fail because the files are hard to build. It fails when nobody owns the index, so the master files drift. Pick a person and put their name at the top of the index.
Every AI job you run sits on top of these. If you want a hand building them, book a meeting. The advisory practice is wired for this.
Frequently asked questions
What is context engineering?
Context engineering is the work of giving an AI system the facts, rules, and relationships it needs before it answers a question. The term became official in Gartner’s July 2025 note that declared prompt engineering out and context engineering in. Underneath the new vocabulary are older ideas like semantic layers and master data.
How is context engineering different from prompt engineering?
Prompt engineering writes a single input that gets an AI to behave. Context engineering builds the environment the AI reads from: positioning, rules, master files, field corrections. Prompt engineering is a skill one person has. Context engineering is a system an organization maintains.
What’s the difference between a semantic layer and a master-file index?
A semantic layer defines what terms like “customer” or “revenue” mean across an organization’s data. A master-file index names the single current file for each fact your AI jobs need. The semantic layer is for data teams. The master-file index is for marketing teams.
How is the marketing team’s context stack different from the data team’s?
The data team’s context stack is a semantic layer plus master data: products a chief data officer buys from vendors. The marketing team’s stack is a positioning doc, a voice guide, a field-correction log, a master-file index, and a buyer-committee map. No vendor sells it; you build it.
Where should a PMM start building a context stack?
Start with the master-file index. For every fact your AI jobs read, pick the single file that holds the current version and mark the others as archive or derivative. Add the field-correction log next, filled in by rep corrections over time.
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 10 AI 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]Three Data Faces Podcast interviews, 2026: Scott Taylor (”The Data Whisperer”), Episode 35, April 7, 2026, https://tinytechguides.com/blog/truth-before-meaning-the-three-word-fix-for-data-management/; Steve Wooledge (CEO, Collate), Episode 39, May 19, 2026, https://tinytechguides.com/blog/why-ai-agents-require-a-switzerland-approach-to-metadata/; Malcolm Hawker (chief data officer, Profisee), Episode 49, September 8, 2026, https://tinytechguides.com/blog/data-faces-malcolm-hawker-ep49-ai-ready-data/.
[2]Gartner. “Lead the Shift to Context Engineering as Prompt Engineering Fades.” Gartner First Takes, July 2025. https://www.gartner.com/en/documents/6781234

