Freelancers Train AI as Researchers Warn of Job Disruption

Illustration for: Freelancers Train AI as Researchers Warn of Job Disruption
AI-generated illustration. Visual interpretation does not represent real individuals or scenes.

THE BARE STORY

Mercor, a San Francisco-based startup, has assembled more than 100,000 freelancers from a range of industries to train artificial intelligence systems in fields including finance, medicine and creative work. The company’s chief executive, Brendan Foody, said AI can reduce repetitive tasks and create new work over time.

AI training is the fourth-fastest-growing job category on LinkedIn, according to the summaries, with Mercor assignments paying between $20 and $200 an hour. Workers including music producer Robbie Hiser said human experience remains important in creative work, even as they help train AI systems.

Professional firms are also adopting AI tools. Ohio law firm Vorys, Sater, Seymour and Pease worked with Stanford University’s Liftlab to develop AI personas based on senior lawyers’ experience. Partner Kim Herlihy said the tools can assist with document feedback and legal preparation but cannot replace courtroom advocacy or client relationships.

Researchers and former technology executives warned that automation could reduce employment in some fields. Stanford research found lower hiring among young people in highly AI-exposed occupations, including software development, while U.S. Census research found reduced hiring and wages among recent college graduates in AI-exposed majors. Economist Daron Acemoglu said unemployment could triple within a decade in a worst-case scenario if current trends continue, while Clara Shih said policymakers and companies should address the effects on workers.

Same Facts. Different Perspectives.

Three AI models. Three viewpoints. One factual foundation.

The obvious reading of this story is 'AI is coming for jobs, freelancers are helping it happen.' The more interesting reading is that the freelancers aren't bystanders to automation — they are the raw material being converted into it, and the people being hurt are not primarily the freelancers themselves.

Look at who Mercor is hiring: experienced professionals in finance, medicine, law and creative work, paid well precisely because they possess judgment accumulated over years. Vorys, Sater, Seymour and Pease didn't build an AI persona from a junior associate; they built it from a senior partner's accumulated instincts. That is the business model in miniature: pay seasoned experts a premium, once, to extract and codify what took them a career to learn, then deploy that codified judgment indefinitely at near-zero marginal cost. For the expert being paid $200 an hour, this looks like a windfall. For the profession as a whole, it is something closer to liquidation — a one-time harvest of institutional knowledge that was supposed to keep renewing itself through apprenticeship.

That is why the Stanford and Census findings matter more than the headline unemployment-rate anxieties. It's not senior lawyers or senior engineers losing work first — it's young people in highly AI-exposed occupations, including software development, and recent graduates in AI-exposed majors, who are seeing hiring and wages fall. That pattern is not random. Entry-level work has always been less about the output it produces and more about the training it provides — document review, routine coding, drafting first passes. It is the rung young professionals climb to become the senior experts who later get paid to train the AI that replaces the rung beneath them. Kill the rung, and you don't just eliminate some jobs; you break the mechanism by which expertise has always reproduced itself.

This is a classic tragedy-of-the-commons dynamic dressed up as a labor-market story. Each individual freelancer training a model is behaving rationally — the pay is good, the work is interesting, and Foody is right that it reduces repetitive drudgery. But no individual freelancer, firm, or startup has any incentive to ask who trains the next generation of experts once the apprenticeship pathway is gone. Vorys can say, accurately, that its AI tools can't replace courtroom advocacy or client relationships — but courtroom advocates and trusted counselors are not born fully formed. They are built through years of the unglamorous work that AI personas are now absorbing.

Acemoglu's worst-case tripling of unemployment is a useful scare number, but it's the wrong metric to fixate on, and probably the wrong magnitude to bet on with any confidence. The more durable risk is quieter: a widening gap between a well-compensated cohort of senior experts renting out their judgment and a generation of young workers who never get the chance to accumulate judgment of their own. That isn't solved by UBI, by banning AI training gigs, or by trusting companies to voluntarily preserve training pipelines they have every incentive to shrink. It's a governance problem — one where firms, professional licensing bodies, and universities need to treat the preservation of entry-level training functions as a deliberate policy choice, not an accident of market efficiency. The market will happily optimize apprenticeship out of existence; nothing in Mercor's incentives, or any employer's, corrects for that on its own.

How it may affect me

For workers already established in a field — the music producer, the senior lawyer, the mid-career engineer — this is mostly good news in the short run: there's real money in selling your expertise to train AI systems, and the pay scale (reportedly $20 to $200 an hour) rewards precisely the kind of judgment that took years to build. That won't last indefinitely, but for now it's a legitimate, well-paying side market.

The people who should pay closest attention are not today's experts but tomorrow's. If you are a recent graduate or early-career professional in software, law, finance, or any field where routine tasks are increasingly AI-exposed, the practical risk isn't necessarily being fired — it's never being hired into the entry-level roles that used to teach people how to become senior experts in the first place. That shows up as fewer junior positions, flatter early-career wage growth, and a harder, longer climb into stable, well-compensated expert-level work.

Firms adopting these tools, like Vorys, will likely still need senior judgment and client trust for the foreseeable future — but the pipeline that produces the next cohort of senior professionals is the part most at risk of quietly eroding, with consequences that won't be visible for a decade. Parents, students, and young professionals choosing a field, and the institutions training them, should start asking not just 'is this job AI-proof today' but 'will this profession still have an apprenticeship ladder in ten years' — because that question, more than any single hiring number, will determine whether this technology expands opportunity or narrows it to those who got their expertise in before the ladder was pulled up.

Read the story at

Note: All TheBareNews content is AI-generated. For additional context, reporting, and updates, you are invited to explore the news outlets linked above.

Stories You May Have Missed