AI Labs Pay $100/Hour for Real Work Samples as Industry Drama Intensifies

The AI industry's latest power struggle—where former startups, billion-dollar seed rounds, and allegations of misconduct collide—reveals a sector where personnel moves feel more like HBO scripts than tech news.

Barret Zoph and Luke Metz, both former employees of OpenAI, were recently rehired by the lab after leaving to co-found Thinking Machines Lab. The company described the departures as stemming from "discussions and misalignment on what the company wanted to build—it was about the product, the technology, and the future." OpenAI claims the rehires were planned for weeks, while Thinking Machines alleges Zoph shared confidential information with competitors.

Meanwhile, AI labs are paying $100/hour for professionals to generate training data for agents via controlled environments.

Companies like Handshake, Mercor, Surge, and Turing supply workers who create realistic work samples—such as drafting emails, analyzing spreadsheets, and coding—within these environments. The data is scrubbed for sensitive information before being used to train AI systems.

These environments differ from traditional enterprise software training in their focus on generating synthetic workloads rather than teaching human users.

While enterprise training emphasizes skill development, AI labs use these environments to simulate knowledge work at scale, creating datasets that reflect real-world professional tasks.