Job Seekers Find a New Source of Income: Training AI to Do Their Old Roles
Buzzy AI startup Mercor employs tens of thousands of white-collar contractors in gigs open to anyone with expertise in their own field
One of the Bay Area’s fastest-growing startups is on a hiring spree—with a twist. To get the job, you have to help train artificial intelligence to eventually perform work as well as, or better than, you do.
That company is Mercor, a $10-billion AI startup that has quietly become a major employer in the emerging “AI training” economy. Instead of driving for Uber or delivering food, tens of thousands of highly educated contractors now spend their days reviewing, correcting, and improving the outputs of large language models that power chatbots and other AI systems.
Mercor is not looking for entry-level workers. Its contractor pool includes astronomers, psychologists, engineers, filmmakers, writers, comedians, lawyers, venture capitalists, and medical specialists. A dermatologist can earn up to $250 an hour helping build healthcare decision-support tools. Poets who refine AI’s understanding of rhythm, metaphor, and emotional nuance can earn up to $150 an hour.
In 2025 alone, Mercor hired more than 30,000 contractors to work on projects for some of the world’s largest AI companies, including OpenAI and Anthropic.
This surge comes at a precarious time for the white-collar workforce. With economic uncertainty, tariffs, and companies hesitating to commit to long-term hiring as AI evolves, unemployment in November reached its highest level in four years. Many professionals are sending out hundreds of applications without success. For a growing number of them, Mercor has become a lifeline.
The company offers a $250 referral bonus, which has flooded LinkedIn with job posts and fueled online speculation that Mercor must be a scam. It is not. But the work comes with an uncomfortable irony: many contractors are paid to help automate the very jobs they are struggling to find.
Katie Williams, a 30-year-old video editor in Houston, applied to Mercor after seeing a posting for video editing work. Her interview was conducted by an AI proctor with a male-sounding voice. She was unsure what the role would involve, but at up to $45 an hour, it was hard to pass up.
Six months later, Williams spends her time watching videos, writing detailed captions of what happens in each clip, and rating AI-generated footage. The work is steady, but emotionally complicated.
“I joke with my friends that I’m training AI to take my job someday,” she says.
Her coworkers express similar feelings in Mercor’s Slack channels. Most dislike the idea of replacing themselves—but many feel they have few alternatives.
Mercor’s leadership disputes the bleak framing. In a written statement, a company spokeswoman said many experts view their work as a responsibility: a way to ensure AI systems reflect accurate, thoughtful, and domain-specific knowledge rather than shallow automation.
To get hired, applicants go through an AI-run interview, and some are required to share their screens while demonstrating their skills. Once on a project, contractors must use time-tracking software to ensure they are not outsourcing their evaluations to AI tools—something Mercor says it has caught people attempting to do.
Peter Valdes-Dapena, a 61-year-old automotive journalist, was laid off in 2024 after more than two decades at the same job. Freelance work has been inconsistent, and full-time roles have been elusive. When Mercor appeared in his LinkedIn feed, he applied.
Now he spends 20 to 30 hours a week critiquing AI-generated news articles. He finds the work intellectually demanding—and surprisingly helpful.
“It’s actually improved my own writing,” he says.
Still, he is conflicted. He believes journalism will always need human voices, but he worries that AI will eliminate many jobs.
“I didn’t invent AI and I’m not going to uninvent it,” he says. “If I stopped doing this, would that change anything? No.”
Laura Kittel, an academic searching for nonprofit and government work, also tried Mercor—but was troubled by the contract. She interpreted it as granting royalty-free rights to her existing and future academic work if it could benefit Mercor or its clients.
“I thought that was an overreach,” she says.
When she tried to amend the terms, Mercor’s AI assistant told her the contract was non-negotiable. She could either accept it or walk away.
“It felt like having a lot of your dignity taken away,” she says.
Mercor says the contract only applies if a worker chooses to use their own writing, code, or intellectual property in a project. Anything created independently and not submitted remains the worker’s property.
For Sara Kubik, a solo attorney, Mercor provides useful supplemental income. She has worked on multiple AI-training projects, including one for OpenAI, and finds the work both challenging and eye-opening.
She rejects the idea that AI will replace lawyers, though she expects some support roles to disappear. And while people sometimes accuse her of helping eliminate jobs, she says the reality is less dramatic.
“Training AI has taught me how limited it still is,” she says. “There’s a lot more hype than capability.”