Your voice is now training data. Not metaphorically. Literally.
If you took a Mercor gig at $80-$200 an hour to "share your domain expertise" with a frontier lab, your reasoning patterns, your case studies, your dissent moves, your client-saving instincts — they are now weights inside a model you will rent back from someone else's API. The check cleared. The leverage transferred.
That is the cost: a one-time payment for a permanent asset that compounds on someone else's balance sheet. Meanwhile the question almost no consultant asks out loud is the only one that matters — what would it look like if your expertise generated revenue while you slept, on infrastructure you owned, with terms you set?
I'm Matt Cretzman. I build Knowledge Delivery Systems for experts. I've spent the last three years watching this exact extraction play out in real time, and the recent Mercor data exposure isn't the scandal — it's the receipt. Here is what the leak actually teaches, and the system I built so you never have to learn it the hard way.
The Leak Is Not the Story. The Business Model Is.
When a contractor list, internal prompts, or expert submissions surface from a knowledge-extraction platform, the headlines focus on security. That misses the point. The breach is a surface event. The structure underneath is the story.
Here is the structure: a platform recruits credentialed experts — doctors, lawyers, supply chain veterans, senior engineers, M&A specialists. It pays them by the hour. Experts answer prompts, write rubrics, grade model outputs, and dump tacit knowledge into a queue. That queue feeds a training pipeline. The pipeline feeds a model. The model gets sold back to the market — including, eventually, to the clients those same experts used to serve.
Mercor's CEO has been candid about this on the record. The company's pitch to AI labs is straightforward: we will get you the experts, we will pay them hourly, we will deliver the data. It is a labor arbitrage on cognition. And it works because most experts have never been offered a different deal.
The Mercor model is extraction economy #2 in the framework I use throughout On Whose Terms. Extraction #1 was the silent one — LLMs ingesting the open web, including the books, papers, and posts experts wrote over careers, with no consent loop. The UK copyright fight in 2024-2025 made that explicit: 88% of surveyed creators wanted protection, the government punted, the training continued. Extraction #2 is louder but cleaner-looking, because money changes hands. A check feels like consent. It isn't. It's a transaction with terms you didn't write.
Extraction #3 is the one almost nobody is building toward. Expert-owned. Your IP, your interface, your revenue, your distribution. Your terms.
The Jacob Trade
There is an old story about a man named Jacob who sold his birthright for a bowl of stew because he was hungry in the moment. The hourly extraction economy is that story at scale.
$150 an hour feels like a great rate when you are between engagements. It is. For one hour. The problem is that the asset you are selling — twenty years of pattern recognition compressed into how you answer a hard question — is not an hourly asset. It is a compounding one. Sold by the hour, it produces income once. Sold as a system, it produces income for a decade.
The consultants I respect most are some of the worst offenders here, because they have been trained by their entire industry to think in billable hours. The hour is the unit. The hour is the ceiling. When a platform shows up offering more hours at a higher rate with no client management overhead, it looks like a gift. It is a trade. You are trading the long compound for the short check.
I'm not arguing nobody should ever take this work. I'm arguing you should know what you're trading and you should have an alternative running in parallel.
What an Expert-Owned KDS Actually Is
A Knowledge Delivery System is not a course. It is not a chatbot. It is not a newsletter funnel.
A KDS is a structured representation of your expertise — your frameworks, your decision trees, your case patterns, your scoring rubrics — wired into interfaces that other systems can call. The expert owns the asset. The expert sets the terms. The interfaces are what make it real.
That last part is where MCP comes in. The Model Context Protocol launched in November 2024 as an open standard for connecting models to tools and data sources. In December 2025 it was donated to the Linux Foundation under the Agentic AI Foundation. That sequence matters. It means MCP is not a one-vendor bet. It is becoming the plumbing layer for how AI systems talk to external knowledge.
For an expert, MCP is the door. It is the standardized way an enterprise's AI agent — running on whatever model the enterprise has standardized on — can call into your knowledge system, get a structured answer, log the interaction, and pay you for it. Not training data. A live query against an asset you own.
The stack looks like this:
1. Capture. Your IP gets pulled out of your head and your old decks into a structured store — frameworks, patterns, decision logic, examples.
2. Refine. The raw capture gets shaped into queryable units. This is the work. It's where most experts stall, which is why I built Skill Refinery to do it for them.
3. Interface. MCP-style endpoints, plus a chat surface, plus an embed for your site. Same underlying knowledge, three doors.
4. Meter. Every query is logged. Every client is identified. Every interaction is billable or attributable.
5. Distribute. Your KDS shows up where your buyers already are — inside their AI tools, inside their workflows, on your domain.
The Mercor model has steps 1 and 2 — and stops. The training set is the product. The expert is the input.
The KDS model has all five — and the expert is the owner.
The Numbers That Should Decide This for You
Let me put real arithmetic on the page.
A mid-career consultant doing 20 hours a month on a knowledge-extraction platform at $175 an hour earns $3,500 a month. $42,000 a year. Cap it at three years before the model surpasses the human contributor's marginal value, which is roughly the trajectory frontier labs are on. That is $126,000 in lifetime earnings on that asset, after which the asset is worth zero to the platform and the platform owes you nothing.
The same consultant with a KDS charging enterprise clients $2,000 a month for metered access — five clients in year one, twelve in year two, twenty-five in year three — produces roughly $10,000, $24,000, and $50,000 a month at those steps. Different math. Different category. And the asset still belongs to you in year four.
I am not picking favorable numbers. The KDS path is harder upfront. It requires the refinement work. It requires showing up as a builder, not a contractor. But the slope is the slope, and slope compounds.
The Builder-to-Builder Read
If you are reading this and you have already taken Mercor-style work, you have not done anything wrong. You answered a market signal. The signal was real. The terms were not yours.
The move now is not guilt. The move is to start building the version where the terms are yours, in parallel, this quarter. You don't quit the hourly work on day one. You build the asset alongside it, and you let the asset grow until it dwarfs the hourly line. Then the hourly work becomes a choice instead of a dependency.
This is stewardship language and I'll own it. Your expertise is not just an income stream. It is a gift you have been refining for a couple of decades. How you deliver it — and on whose terms — is a question of integrity, not just strategy. I don't think the experts I work with are in this moment by accident. I think there is a window, and the window has a shape, and the shape is the KDS.
What I'd Do This Week If I Were You
Three concrete moves.
One: inventory what you've already given away. Pull every platform contract, every NDA, every "data labeling" agreement you've signed in the last 24 months. Read the IP and training-use clauses. You will be surprised. Some of you signed perpetual, irrevocable, sublicensable rights for $90 an hour. Know your starting position.
Two: pick one framework to externalize. Not your whole brain. One framework. The one clients ask about most. The one you've taught fifty times. Write it down as a structured decision tree, not a blog post. That is the seed of your KDS.
Three: stop treating MCP as a developer concern. It is your distribution layer. The enterprises you want as clients are wiring MCP into their internal AI stacks right now. If your knowledge has an MCP endpoint by the time their procurement team is shopping for vertical expertise, you are on the shortlist. If it doesn't, you are not in the conversation.
That third one is the part most consultants will get wrong, because it sounds technical. It isn't. It is a packaging decision. The technical work is what I do. The decision is yours.
Closing
The Mercor leak is a lighthouse. It is lighting up exactly where the rocks are. Every expert who reads it correctly walks away from the hourly extraction economy and starts building the owned one. Every expert who reads it as a security story misses the lesson and signs the next contract.
I'm building the infrastructure for the second group to become the first. That's what Skill Refinery is. That's what Stormbreaker Digital wires into enterprise stacks. That's why I write at mattcretzman.com — to give experts the full picture before they sign the next agreement.
I'm writing a book about this — On Whose Terms: The New Expert Economy and the Fight for What You Know. If the thesis resonates, join the launch list.
Your voice is already training something. The only open question is whose model and whose terms.
Keep Building,
— Matt