A startup called Mercor is worth $10 billion. It pays its experts $85 an hour.
If you haven't heard of Mercor, you're about to. It's the marketplace where OpenAI, Anthropic, Meta, Microsoft, Google, Tesla, and Nvidia go to rent the minds of doctors, lawyers, bankers, scientists, and PhDs — so those experts can train the AI models that are coming for every profession those experts spent their lives building.
The numbers, from TechCrunch and Bloomberg reporting: 30,000+ experts on the roster. $1.5 million a day in contractor payments. Founders who are 22 years old. A $10 billion valuation in under three years. A $500 million annualized revenue run rate.
The founders didn't set out to build this. They started a resume-screening company in 2023. They pivoted when they noticed the real opportunity: AI labs were desperate for humans with deep domain expertise to train their models, and nobody had built the marketplace for it.
At TechCrunch Disrupt last year, CEO Brendan Foody revealed the part that should have ended the dinner party. Instead of signing expensive data licensing contracts with Fortune 500 companies, AI labs now go through Mercor to hire former senior employees of those same companies as contractors — and extract the industrial knowledge directly from them. No negotiation with the old employer. No royalty to the institution that developed the know-how. Just the expert, the hourly rate, and the model getting smarter on what used to be proprietary.
One contractor, quoted by Bloomberg: "It's dehumanizing to mentor AI on a 30-year body of work, knowing you're feeding the thing that will bite you later."
Others describe the experience as being "treated like cattle" — intrusive surveillance, legal questions about whether this is even independent contracting or misclassified employment.
This is the piece I didn't want to write. But it has to be written, because it exposes a deep deep problem that I've been trying to explain for over a year, and that I know Skill Refinery solves.
The math that should end the debate
Mercor pays experts $85 an hour on average. At the top end, PhDs and JDs and Wall Street veterans get $150-$200.
Sounds decent. It's not.
Let's do the math an expert should be doing before they click "accept" on the next Mercor assignment.
A senior consultant with 20 years of experience has roughly 40,000 hours of expertise compounded into their professional judgment. Their methodology, their frameworks, their pattern-matching on edge cases — all of it earned, all of it hard-won, none of it replicable by reading a book. They log into Mercor. They get matched with an AI training task. They spend three hours refining a model's reasoning on a problem only someone like them could refine. Mercor pays them $450.
That $450 transfers a piece of their 40,000-hour body of work into a model that will be used — at scale, around the clock, forever — to answer questions that used to require calling that consultant. Let's be conservative and say the model, once trained, handles 10,000 queries per year that would have otherwise gone to a human expert. At $500 per query (which is a low estimate for senior consulting work), that's $5 million in annual service revenue, absorbed by the AI lab, captured by the AI lab's customers, and not shared with the expert who trained it.
The expert got $450. The AI lab got a compounding asset worth millions over its operational life. Mercor got its finder's fee. The lab's customers got a replacement for the expert's billable hours at a fraction of the cost.
This isn't a marketplace. It's an extraction.
And the worst part — the part that keeps me up — is that the experts are doing this voluntarily. They're signing up. They're taking the gigs. Because they see no other option.
Why they have no other option
Let me be fair to the experts for a second. They're not stupid. They know what they're doing.
Many of them are watching their own industries get disrupted. A radiologist knows AI is coming for image interpretation. A paralegal knows AI is coming for discovery. A financial analyst knows AI is coming for research. A copywriter knows AI is already here.
The Fortune/Duke CFO survey from March 2026 projected 502,000 AI-attributed layoffs this year — a 9x increase over 2025. Block cut 4,000 people in March and the CEO wrote the memo himself: "This is not driven by financial difficulty, but by the growing capability of AI tools." Oracle cut 10,000+. Anthropic's own CEO, who pays Mercor's invoices, is on record predicting half of entry-level white-collar jobs gone in 18 months.
So when a platform offers $85-$200 an hour, flexible schedule, no boss, work from home, use the expertise you already have — of course experts take it. At least they're getting paid something. Most of them have already watched AI models trained on their public work get deployed with no compensation at all. The New York Times is suing OpenAI. Sarah Silverman is suing Meta. The U.S. Copyright Office issued a report saying AI training on copyrighted work is not inherently transformative. None of that has put money in the pocket of a single expert whose book was ingested into a foundation model.
Mercor, at least, pays.
That's the bar. At least they're getting paid. That's how low we've set the floor for experts participating in the AI economy.
The asymmetry nobody is naming
Here's what's actually happening in that transaction, stripped of the marketing language.
The AI lab has a problem: their model is mediocre at a specific kind of expert reasoning. Hiring a full-time expert costs $300K a year plus benefits. Licensing a body of knowledge from a corporation costs millions and requires lawyers. But renting the expert by the hour through Mercor costs $85-$200 and comes with no ongoing liability.
The expert has a different problem: their industry is compressing, their billable hours are getting harder to justify, and they don't know what their career looks like in five years. The Mercor offer is real money, today, for work they can do at 9 PM after the kids are in bed.
Both sides rationally agree. The transaction happens. The model gets better.
But look at what's actually being exchanged. The AI lab pays for three hours of expert reasoning. The expert provides three hours of expert reasoning. That's the surface transaction.
Underneath, the AI lab has acquired an irreversible capability. The model learned from that expert. The model now has that piece of expertise baked in, forever, for every future query, at zero marginal cost. Every person who previously would have paid for that expert's time — every client, every employer, every downstream professional service — can now be served by the model the expert helped train, at a 95% cost reduction.
The expert sold a compounding asset for a one-time fee. The AI lab bought an asset that compounds. They are not the same transaction.
And here's the brutal part: even if the expert knew this and wanted to negotiate differently, they couldn't. There's no mechanism. Mercor's pricing is take-it-or-leave-it. The AI labs don't offer royalties. The expert has no legal or technical way to say: "I'll train your model, but I want a share of the revenue every time my expertise gets invoked in a customer query for the next five years."
That mechanism doesn't exist.
Until it does.
Why this is so wrong
Let me be precise about what I think is wrong here, because it's important.
I don't think Mercor is evil. The founders built a real marketplace that solved a real problem for the AI labs. The experts get paid, which is more than most experts have gotten from the AI boom so far. Mercor isn't the villain of this story. It's the evidence.
What's wrong is the asymmetry of the structure. An expert's life work is their most valuable asset. For most professionals — the doctor, the lawyer, the consultant, the accountant, the engineer, the coach — it's the only asset they have that's truly theirs. They built it. They earned it. They paid for it in the most expensive currency there is: the hours of their life.
And the only marketplace that currently exists for monetizing that expertise in the AI economy is one that takes permanent ownership of it for a one-time payment. That's not a market failure. That's a missing market.
A plumber doesn't sell his plumbing van to the company he works for in exchange for an hourly rate that's marginally better than what he made before. A brand doesn't license its entire brand identity to a competitor for a modest annual fee. A musician doesn't hand over their master recordings to a label for a few hundred dollars per session. All of those industries figured out, over decades, the difference between labor and equity, and they built the legal and financial structures to keep experts from accidentally giving away the wrong one.
We haven't done that for knowledge workers in the AI era. Yet. We're in the early stage where experts don't know the difference, and the companies buying know exactly what they're doing. The Mercor model is built on that asymmetry.
That's what's wrong.
What the alternative looks like
I've spent the last eighteen months building Skill Refinery on the belief that this asymmetry isn't permanent. That there's another way for an expert to participate in the AI economy — one where they keep ownership of their expertise, get paid every time it's invoked, and don't have to hand over a compounding asset for a gig rate.
Here's how it actually works.
You have a body of knowledge — books, frameworks, client playbooks, the methodology you've refined over twenty years. Instead of renting yourself by the hour to train somebody else's model, you extract that knowledge into skill cards under your name. Each card is scoped to a specific problem. Each card lives in your library. Each card is invoked by AI agents — ChatGPT, Claude, Copilot — when someone needs your specific expertise.
The revenue model is a subscription marketplace. Subscribers pay $200 a month for library access. The expert earns $80 per attributed subscriber, $60 per organic subscriber. A single established body of work can produce $3,600 to $72,000 a month in recurring revenue depending on tier.
Compare the two structures.
Mercor: expert sells three hours of expertise for $450, one-time, no future participation. AI lab absorbs it permanently and earns indefinitely from the resulting model improvement.
Skill Refinery: expert publishes skill cards once, earns recurring royalties every month those cards are invoked, retains attribution and ownership, and accumulates a compounding income stream that grows with agent adoption.
Mercor's model is structurally rigged toward the platform and the AI labs. Skill Refinery's model is structurally rigged toward the expert. That's not a value judgment about Mercor — it's just a description of how the money flows.
The reason this matters is that experts right now are being told there are only two options: fight AI, or rent yourself to AI. Neither of those actually works. Fighting AI fails because the models are already trained on your public work. Renting yourself to AI fails because the rate doesn't compound.
There's a third option. Own your skill cards. Get paid every time an AI invokes them. Let the agent economy work for you instead of on you.
What I'm asking experts to do
Read the Mercor coverage. All of it. Read the contractor who said it felt dehumanizing to mentor AI on a 30-year body of work. Read the Bloomberg reporting on the surveillance. Read Foody's TechCrunch Disrupt comments about extracting industrial knowledge from former senior employees.
Then ask yourself the honest question: is this the deal I want for my life's work?
If the answer is no, stop signing up for Mercor gigs. Stop feeding the thing that will bite you later. The money is real. The cost is irreversible.
If the answer is "yes, but I don't see another option," I'm telling you there is one. It's not theoretical. Skill Refinery is a live platform with experts already using it to publish skill cards into ChatGPT, Claude, and Copilot under their own names, with attribution, with royalty structures, with recurring revenue.
You built your expertise over decades. The next decade is the one that decides whether you own it or whether somebody else does. That decision is yours for a narrow window, and then it isn't.
Don't sell your life's work by the hour. Claim it.
Matt Cretzman operates Stormbreaker Digital as a fractional CMO and fractional Chief AI Officer. He's the founder of Skill Refinery, the platform for experts who want to own their expertise in the AI era — not rent it out.
Keep Building,
Matt