When the people training AI are training themselves out of a job

Every day, a three-year-old start-up called Mercor pays more than 30,000 contractors upward of $4 million to help make their own jobs, and the jobs of their colleagues, obsolete. That figure was first reported by The New York Times in July 2026, and it says a lot about where this business has gone in three years. The contractors on Mercor's books aren't clicking through simple image-tagging tasks. Recent postings have called for a voice actor who can hold a customer-service persona in fluent Hebrew, a physicist specialising in general relativity, and a physician with specific experience in the Rwandan primary care system.

This is the strange new frontier of white-collar gig work. Highly credentialed professionals, lawyers, doctors, PhDs, consultants, are being paid well to hand over the judgement, phrasing, and decision-making patterns that took them years to build, so AI models can learn to reproduce them. For HR leaders it's not just an interesting news story. It previews how expertise is being unbundled, and a reminder that an organisation's most valuable knowledge can be extracted, packaged, and sold, sometimes by the very people who built it.

A gig economy for the credentialed class

Mercor was valued at $10 billion in an October 2025 funding round and, as of July 2026, was reportedly in talks with investors for a valuation double that, according to The New York Times. It isn't alone. Meta invested more than $14 billion in rival Scale AI last year, in part to bring on its chief executive. Handshake, a recruiting start-up that pivoted into data training, said its annualised revenue run rate crossed $1 billion in April 2026, up from $550 million at the start of the year.

The work itself has shifted toward higher-credentialed, more specialised expertise. Earlier data-labelling operations relied on low-paid overseas workers tagging images. Today's AI labs need mathematicians to annotate formal proofs, lawyers to mark up legal briefs, professors to grade essays. The open internet has largely been scraped already, so further gains now depend on capturing the tacit expertise that never made it online in the first place.

Some firms are going further still, trying to recreate entire workplaces in miniature. Mercor's founder, Brendan Foody, described the acquisition of Deeptune, a start-up that builds simulated versions of workplace software like Slack and Salesforce, as solving "the bottleneck to a frontier lab automating everything that people do." In practice, that means pairing role-players as clients and employees to model exactly what happens inside, say, an investment bank.

What the work actually looks like

The daily reality of these jobs is less glamorous than the funding rounds suggest. Reporting by The Guardian in 2025 described AI "raters" contracted through Hitachi's GlobalLogic to evaluate and moderate output from Google's Gemini and AI Overviews. One rater described a task timer that shrank from 30 minutes to 15, in which she was expected to read, fact-check, and rate roughly 500 words of AI response each time. Rachael Sawyer, a Texas-based technical writer who has rated Google's AI products since March 2024, said she was expected to complete dozens of such tasks a day, each within a 10-minute window, and that the job pulled her into content moderation, including flagging violent and sexually explicit material, that was never mentioned in her onboarding or job description. Researcher Adio Dinika, who studies the human labour behind AI at the Distributed AI Research Institute, put it starkly: "AI isn't magic; it's a pyramid scheme of human labor."

The professionals doing more specialised work described a similar arc. Amanda Brown, an assistant professor of biology at Tarleton State University in Texas, took on gigs through Mercor and Handshake and told The New York Times the pay looked appealing at first, but the hours ballooned, deadlines tightened, and feedback from supervisors felt demoralising. Carolina Perez Sands, a speech and language pathologist who trained language models in Portuguese and creative writing for Mercor, described a similar trajectory to The Wall Street Journal's "The Journal" podcast in June 2026. Her corrections became unnecessary within about a week as the model absorbed them, and she eventually concluded that "my job is actually making this more of a monster." She has since left the industry for a sales role.

Workers training themselves out of a role within months came up across multiple accounts. It's also, by design, the business model. These firms need AI models to keep improving to prove their value to clients, but they need those same models to stay imperfect enough that clients keep coming back for more data.

The economist's warning

Writing in the Financial Times in March 2026, MIT Sloan management professor Danielle Li argued that this dynamic threatens something more fundamental than any single job. "Historically, economic security has rested on the scarcity of skill," she wrote. Once a top performer's judgement is codified into a model, she noted, it can be distributed to every other worker in that role, everywhere, far faster than any human mentorship chain could manage.

Li's advice to workers comes down to three shifts in thinking. On productivity: if your work trains a model that benefits your employer, seek recognition and pay for it rather than assuming your normal wage already covers it. On competition: AI turns what used to be a local labour market for expertise into a global one, since anyone able to produce similar work can now help build a model that competes with you directly. On cooperation: workers who supply training data cheaply and individually can undercut their own collective bargaining position without realising it, which is why she argues coordination among workers may matter more than any individual career move.

Not everyone in the industry agrees on how long the current gold rush will last. Anton Korinek, an economist on leave from the University of Virginia to work at Anthropic, told The New York Times that both the importance of post-training data and the growth of companies that supply it may ease somewhat as models improve in a given domain. That's a more measured view than Surge AI founder Edwin Chen's, who has predicted training AI models could become one of the most respected and consequential career paths in society.

What this means for HR

For HR functions, the data-training economy raises questions that go well beyond the fate of one gig-work sector.

The first is that your own workforce is already generating training data, whether anyone has thought about it or not. Support tickets, sales calls, code repositories, and case files are all potential inputs for internal AI tools, and per Li's framing, that shifts the definition of "doing your job" to include a second, unrecognised output: the data trail it leaves behind. That's part of why connecting skills data to broader workforce systems is becoming central to HR's AI strategy. Organisations that get ahead of this now will be better placed to have an honest conversation with employees about how their expertise gets used.

The second is that AI training can no longer be left informal. A recent Express Employment Professionals/Harris Poll survey found that 79% of Canadian job seekers want structured, employer-led AI training rather than being left to trial and error, and 77% of hiring managers agree it should be a company priority. Employers who leave that gap unfilled risk employees picking up gig-economy habits toward their own expertise on the side, with little visibility into what's actually being shared.

The third is that retraining plans need to account for people whose roles may not survive intact. Microsoft's approach to cross-training entry-level legal staff, rather than simply eliminating roles as AI reshapes that function, is one model for how large employers are trying to get ahead of the disruption rather than just reacting to it. That sits alongside the harder question of whether it's worth investing in AI skills for workers who may not stay long enough to repay that investment, a calculus that looks different once you consider the alternative may be losing that expertise to a data-training platform instead.

As Dionne Woo, chief people officer at SiteMinder, told HRD, planning for an entirely unknown future, one where lost roles may not return in familiar form, recently dominated conversation at a gathering of HR leaders she attended. It's part of a broader shift in how AI and analytics are reshaping HR's foundations.

Right now, one of the fastest-growing categories of work involves people getting paid to make their own expertise redundant. Whether that trade is worth it, for the worker or for the employer footing the bill, is a question HR departments are increasingly being asked to help answer.