White-collar workers are getting paid $200 an hour to train AI on their jobs — but they say it’s not 'easy money'

“It seems like easy money,” Brown told The New York Times. “It really isn’t.”

Brown is one of tens of thousands of white-collar professionals now being paid to teach AI systems the jobs they spent years learning. A July 10 Times report put numbers on the trade, showing how Mercor, a San Francisco-based startup that sells training data to AI labs, pays its network of 30,000 contractors upward of $4 million every day.

Mercor was reportedly in early talks in July that would peg its worth near $20 billion, according to Bloomberg.

Whether the work is worth it is a harder question now, and the people doing it have started answering.

What the work pays and who’s buying

The money sounds good. A listing on Mercor’s job board offered up to $350 an hour for a psychiatry expert to build clinical scenarios and check the model’s answers against real medical standards, CBS News reported. Surge AI has advertised $300 to $500 an hour for management consultants and investment bankers, and up to $1,000 for venture capital partners.

Alignerr, a Labelbox subsidiary, pays contractors with credentials like a chemistry doctorate up to $200 an hour.

But most gigs pay less than that. Mercor told CBS News its roles average about $105 an hour, and a generalist job reviewing AI search results for errors pays $50 an hour. Brown herself thought that $60 an hour seemed like an attractive rate for these tasks.

Who’s paying for all this expertise? The AI labs are. The big models have already read most of the internet, so the next round of improvement depends on experts checking their work. That could mean, for example, a law professor grading the model’s brief the way she’d grade a first-year associate’s. “We are now at a stage where they need more fine-tuning and reinforcement,” Christine Cruzvergara, chief education strategy officer at recruiting startup Handshake, told CBS News.

Handshake only moved into AI data work in 2025. It told the Times its annualized revenue rate hit $1 billion in April, nearly double the $550 million pace it started the year at.

The industry insists this is good news for workers. Edwin Chen, the billionaire founder of Surge AI, told the Times his company is creating paid work and building AI collaborators that will help knowledge workers do more rather than replace them, and predicted that training AI will one day rank among the “most prestigious” careers going.

Where the easy money gets hard

Ask the contractors and a different picture comes back. Sommer Wall, a lawyer who contracted with Mercor, told the Times she slept in snatches on her couch to survive three-day deadline sprints on the most demanding projects.

Brown grew frustrated with flat-rate gigs that took far longer than she’d budgeted for. On some projects, per the Times, contractors get paid only if a supervisor approves their output.

There’s also a deeper problem.

Once a group of experts teaches a model a skill, projects in that specialty dry up, and this has been a complaint all over the Discord servers and Reddit threads where contractors compare notes, per the Times.

Carolina Perez Sands, a speech and language consultant who worked with Mercor, told “The Journal,” The Wall Street Journal’s podcast, that the model learned from her fixes so fast that, in a matter of weeks, it barely needed her.