Your judgment is already being packaged. Right now. Some agent somewhere is answering a question only you used to be able to answer, and the person who built that agent paid someone $85 an hour to extract the reasoning behind it.
That someone wasn't you. Or maybe it was, and you signed an NDA so you can't talk about it. Either way, the value of that judgment in the open market just dropped, and the agent's price-per-query is now a rounding error on your old consulting rate.
What would it be worth to own that agent instead of feeding it? To sell the answer at scale, on your terms, with your name on the door?
That's the question Microsoft's 2026 framing forces every expert to answer. And I built a system around the answer. Let me walk you through it.
The Partner Language Is Marketing. The Supplier Reality Is Math.
Microsoft is now calling AI agents "partners, not instruments." It's a tidy reframe. Collaborative. Warm. The kind of language that makes a CIO feel good about a $40M Copilot rollout.
But partnership is a two-sided word. If the agent is the colleague, who's the supplier? Every agent — every single one — runs on judgment it didn't generate. Pattern recognition trained on someone's decisions. Heuristics distilled from someone's mistakes. Frameworks lifted from someone's career.
The agent is a colleague to the buyer. To the expert whose brain it was built from, it's a customer that already paid once and will never pay again.
That's the math. I'm Matt Cretzman, and I've spent the last three years building systems for experts who saw this coming and wanted to land on the right side of it.
The Three Extractions, Updated for the Agent Era
In my book On Whose Terms, I lay out the three extraction economies every expert is now operating inside. Pick one. You don't get to opt out.
Extraction One: LLMs ate your work without asking. The UK copyright fight made this explicit. 88% of surveyed creators wanted protection. The government punted. OpenAI, Anthropic, Google — they ingested books, papers, transcripts, courses, podcasts. Your blog from 2017 is in there. Your conference talk from 2019 is in there. No check arrived.
Those were their terms.
Extraction Two: Mercor and the hourly dump. Mercor's CEO went on Redpoint's podcast in June 2025 and explained the model plainly. Pay experts hourly — radiologists, attorneys, M&A bankers, surgeons — to sit at a screen and explain their reasoning while it's recorded, structured, and routed into a training corpus. The expert gets $80 to $200 an hour. The buyer gets a permanent asset that replaces a class of professionals.
I call this Jacob selling his birthright for stew. The stew is real. It's also stew.
Those are their terms.
Extraction Three: Expert-owned. You package your own judgment. You build the agent. You own the cards, the revenue, the distribution, the customer relationship. The buyer rents your reasoning by the query, not your time by the hour. The asset compounds for you, not someone's Series C.
Those are your terms.
Microsoft's "partner" framing collapses cleanly into this map. Every agent-as-colleague has an expert-as-supplier behind it. The only variable is whether that supplier is you or someone like you who signed faster.
What "Supplier" Actually Means in Dollars
Let's anchor this. A senior radiologist in the U.S. earns roughly $400K. Mercor-style platforms pay that same radiologist $120 an hour to train a diagnostic agent. Forty hours of work. $4,800. The agent then runs 24/7, serves thousands of queries, and is sold into hospital systems at a per-scan price.
Do the unit economics on that. The expert traded a career-defining knowledge asset for one week of side income. The platform built a perpetuity.
Now flip it. The same radiologist builds a Knowledge Delivery System — a structured packaging of their diagnostic reasoning, edge cases, decision trees, and confidence calibration. Sells access to mid-tier radiology groups at $2K a month. Ten customers. $240K a year. Recurring. Owned. No NDA. No platform tax.
Same brain. Different terms. The gap is roughly two orders of magnitude over five years.
This isn't theoretical. I'm running variations of this build right now for experts in compliance, M&A diligence, clinical operations, and engineering hiring. Same pattern every time.
The Knowledge Delivery System, In Plain English
A KDS is four things. None of them are a course.
One: Structured capture. Your judgment gets pulled out of your head through a specific interview architecture. Not transcripts. Not bullet points. Cards. Atomic units of reasoning that can be recomposed, retrieved, and licensed.
Two: An agent layer that runs on those cards. Not a generic chatbot with your name on it. A retrieval system that answers in your voice, with your frameworks, citing your reasoning. The customer talks to your expertise without talking to you.
Three: A distribution channel you own. Email list. Direct sales. Partner network. Not a marketplace that can deplatform you on a Tuesday because some algorithm flagged your category.
Four: A revenue model that compounds. Subscription, not hourly. License, not labor. The asset gets more valuable as more people use it, because usage generates feedback that sharpens the agent.
That's what I built Skill Refinery to deliver. It's the proof-of-category for expert-owned extraction. Not the only way to do this. Just the way I do it.
The Decision Has a Clock On It
Here's what's different about mid-2026 versus mid-2024. The agent infrastructure is now commodity. Anthropic, OpenAI, and the open-source stack have all converged. Building an agent is no longer the hard part. The hard part is the corpus — the proprietary judgment that makes one agent worth paying for and another worth ignoring.
Which means experts with packaged IP have a window. Maybe eighteen months. Maybe less. After that, every category has its incumbent agent, distribution is locked, and the experts who didn't move are competing against an agent built from their peers' brains for a fraction of their billing rate.
I'm not selling fear. I'm reading the chessboard. The Microsoft partner framing is the tell. When the largest software company on earth starts calling agents your colleagues, they're not describing the future. They're describing the present and asking you to feel okay about it.
You can feel okay about it. Or you can build the supplier side on your terms.
The Faith Note
A quick word, then I'll move on. I believe expertise is a gift held in trust. You spent twenty years sharpening it. That sharpening was a stewardship, not an accident. The question of how you steward it next — whether you let it get extracted for stew or you compound it into something your kids inherit — that question is older than AI. The technology just made it urgent.
Decide on purpose. That's all I'll say about it.
What This Looks Like If You're Reading This
You're probably between forty and sixty. You've got a credential, a reputation, a network, and a body of work that other people quote. You've been asked to "consult" for an AI company at a rate that felt good in the moment. You've watched a junior version of your job get automated and thought that won't happen to me — and then noticed it happening to a peer.
You don't need to panic. You need to package. The window is open. The tooling is real. The buyers are paying.
I wrote a book about this — On Whose Terms: The New Expert Economy and the Fight for What You Know. It lays out the three extractions in detail, the KDS framework chapter by chapter, and the specific moves experts are making right now to land on the right side of the math. Grab a copy here.
If you want to see what an expert-owned build looks like in practice, the full system is at mattcretzman.com. That's where Skill Refinery, Stormbreaker Digital, and TextEvidence all connect.
Agents as colleagues means experts as suppliers. The only question left is whose terms you sign.
Keep Building,
— Matt