a gap we kept watching get worse.
A few years ago, AI training was a sleepy corner of the internet, a few platforms, a small pool of contractors, rates that wandered between $8 and $25 an hour. Then the frontier labs started spending real money on training data, and the whole thing exploded. Tens of thousands of trainers, billions of dollars in budgets, and the same fundamental dynamic: the platforms got bigger, the labels got better, and the people doing the actual judgment work kept getting treated like interchangeable parts.
We watched smart, capable people, PhD candidates, working professionals, multilingual specialists, log into Outlier or Scale or Mercor, rate AI outputs for hours, and walk away with $11/hr. Meanwhile the labs they were training were valued at tens of billions. Something was off in the math.
what was missing wasn't work. it was advocacy.
The work itself was there. The demand was there. The talent was there. What was missing was anyone whose job it was to be on the trainer's side. Someone to negotiate rates. Someone to push back when a platform retroactively rejected work. Someone to translate ambiguous rubrics into clear guidance. Someone to track payments across three platforms and notice when something didn't add up.
Every other expert market has those people. Actors have agents. Authors have agents. Athletes have agents. AI trainers, the human substrate underneath the biggest software wave in a generation, had nothing. Just platforms, on the other side of the table.
theNumen is the first attempt we've seen at being the thing on the trainer's side. Not a job board. Not a marketplace. An expert network with a service layer that earns its fee by making trainers measurably better off than they'd be alone.