AI companies hire temp workers to teach chatbots human skills

AI companies are paying writers, doctors, wine hobbyists and other temporary workers to teach chatbots the human skills those systems still lack, creating a labor market built to refine tools that could eventually replace parts of the same workforce. The jobs are increasingly described as some of the fastest-growing in AI, even as they depend on people doing hands-on work that the industry likes to market as automated.

Christine Cruzvergara of Handshake said these AI-training roles are "some of the fastest-growing jobs out there." Her explanation goes to the heart of the boom: large language models have already consumed much of the available data, so the next stage depends on more fine-tuning and reinforcement from humans who can judge outputs, correct errors and supply specialized expertise.

The hiring lists reach far beyond coders. Recruiters are looking for creative writers, air traffic controllers, litigators, improv actors, communications professionals, photo editors, musicians, venture capitalists, doctors, foreign-language speakers, chess champions and wine enthusiasts. The work is often project-based, may require nondisclosure agreements, and postings frequently do not name the AI developer behind the assignment, keeping the labor chain hidden even as the models themselves become more visible.

Mercor, one of the platforms connecting applicants to these jobs, says it links people with AI labs and enterprises to provide the human expertise needed for AI development. The company says the workers it places earn an average of $105 an hour, with some highly specialized roles paying much more. A psychiatry-focused training job listed by Mercor pays up to $350 an hour to design clinical scenarios and evaluate model outputs.

Brendan Foody, Mercor’s chief executive, said "training agents is going to become the largest job category in the world." The claim captures the scale of what is emerging: a new human supply chain behind supposedly automated systems. Robin Palmer, a Hollywood screenwriter and author, spends about 30 hours a week training chatbots to write creatively, and says she sees the models improving.

The contrast with earlier AI labor is stark. Years before today’s hiring surge, OpenAI contractors were earning about $15 an hour with no benefits for behind-the-scenes label-and-feedback work. The pay gap, and the continued reliance on human judgment, shows that even the most advanced AI systems still depend on workers whose labor is often invisible, temporary and essential.