Imagine earning $375 an hour to teach an artificial intelligence system how to do your job – only to realize you might be training your own replacement. This is the unsettling reality for thousands of highly educated professionals working for companies like Mercor, a Silicon Valley startup that pays experts to train AI models in fields ranging from management consulting to radiology. As AI continues its relentless march into knowledge work, a new industry has emerged where skilled workers are paid premium rates to essentially automate their own professions.
The Lucrative Business of Training Your Successor
Mercor, founded by three school friends, has grown into a $10 billion company that pays out about $2 million daily to approximately 30,000 experts. These professionals, many holding advanced degrees, guide AI models through complex problem-solving scenarios, critique their responses, and teach them sophisticated tasks. The average hourly wage exceeds $95, with specialized roles like radiologists commanding up to $375 per hour. “It’s training the LLM [large language models] to do the job,” says Lola, a business graduate who asked to use a pseudonym due to contract terms.
The CEO’s Vision Versus Economic Reality
Mercor’s 22-year-old CEO Brendan Foody, who became one of San Francisco’s youngest billionaires according to Forbes, paints an optimistic picture of AI’s future. He envisions a world where AI handles mundane tasks, allowing humans to focus on higher-level work and “accomplish far more.” Foody acknowledges there will be “some job displacement” but believes new job categories will emerge. However, labor economist Zoe Cullen from Harvard Business School warns that the gig nature of this work leaves trainers unprotected if they help build models that ultimately threaten their jobs. She suggests trainers should retain a stake in revenue produced by models using their expertise.
The Davos Divide: Tech Titans Clash Over AI’s Future
While companies like Mercor build the infrastructure for AI advancement, tech leaders at the World Economic Forum in Davos revealed deep divisions about the technology’s trajectory. Google DeepMind CEO Demis Hassabis warned that parts of the AI industry show “bubble-like” investment patterns, citing multibillion-dollar seed rounds in startups without products or technology. Meanwhile, Microsoft CEO Satya Nadella emphasized the need for broader AI adoption, suggesting that without increased usage, the industry risks a bubble burst. Nvidia CEO Jensen Huang focused on job creation potential, while Anthropic’s Dario Amodei sparked controversy by comparing AI data centers to “a country full of geniuses” in geopolitical debates about chip exports to China.
The Human Cost: Efficiency Versus Employment
The tension between AI efficiency and human employment becomes stark when examining robotics. UBTech, a leading Chinese humanoid robot maker, revealed that its Walker S2 robots are only 30-50% as efficient as human workers in specific tasks like stacking boxes and quality control. Despite this inefficiency, manufacturers are racing to order them to avoid competitive disadvantages. UBTech aims to boost robot performance to 80% of human efficiency by 2027 and targets producing 10,000 robots this year. This push comes as China accounted for more than half of global industrial robot installations in 2024.
Emerging Roles in the Agentic AI Revolution
As AI transforms workplaces, new job categories are indeed emerging, though they may not offset displacement. According to Andie Dovgan, chief growth officer at Creatio, four key roles will lead the agentic AI revolution: AI leaders, agent operators, AI no-code creators, and workflow architects. These positions require a blend of business expertise, AI literacy, and no-code configuration skills. “AI is not simply being added as another layer of automation,” Dovgan explains. “It requires building new workflow architecture. It is reshaping how work itself is designed, executed, and governed.”
The Productivity Paradox and Workplace Isolation
Beyond job displacement, AI introduces subtler workplace challenges. Employees increasingly turn to chatbots for coaching and companionship, seeking what one executive called “the colleague with no drama.” This desire for frictionless interaction, intensified by remote work, risks creating workplace isolation. A study found that social interactions help “idea development” and spread “tacit and informal knowledge” – elements AI cannot replicate. Business psychologist Tomas Chamorro-Premuzic notes that even human executive coaches often prioritize ego-enhancement over hard truths, creating a culture where AI’s reflexive positivity might not be so different from human behavior.
The Economic Impact and Uncertain Future
The AI training industry already contributes significantly to the economy. A study commissioned by Scale AI found the U.S. data annotation industry contributed $5.7 billion to GDP in 2024, projected to rise to $19.2 billion by 2030. Yet economists remain divided about the ultimate impact. Anton Korinek, director of the University of Virginia’s Economics of Transformative AI initiative, warns that if AI becomes transformative enough, “the big issue is what do we do with everybody, essentially.” London Mayor Sadiq Khan has already warned that AI risks creating “mass unemployment” in white-collar sectors unless protective measures are implemented.
Navigating the Transition
For now, many AI trainers embrace their paradoxical position. Jay Katoch, who spends 40-80 hours weekly on Mercor projects after decades in consulting, finds working with AI fulfilling. “You’re challenging [the models] and correcting them,” he says. Others appreciate the high salaries and front-row seats to technological development. As one 18-year-old contractor noted, “Me and the people who work on these products have the experience not to lose their job – people who don’t have that opportunity, I believe they will.” This creates a troubling divide between those training AI systems and those who might be replaced by them.
The AI training industry represents both opportunity and peril in the technological transformation of work. As companies like Mercor create temporary high-paying jobs teaching AI to perform knowledge work, they’re simultaneously building systems that could make many of those jobs obsolete. The question isn’t whether AI will transform work – it’s whether we’re building systems that enhance human potential or simply replace it, and whether the workers training today’s AI will have a place in tomorrow’s economy.