CharlieKey takeaways from your favorite podcasts
← 10 November 2026Sourcery with Molly O'Shea

Mercor Raises $750M at $20B Valuation | CEO Brendan Foody

Brendan Foody argues that as agents automate repetitive work, humans will contribute by teaching models through evals, rubrics and RL environments, making expert data a durable, growing market that Mercor aims to lead.

Watch on YouTube ↗

Charlie's summary

  1. Humans will contribute to the economy by training agents once, not doing repetitive work repeatedly

    Brendan thinks work will shift from repetitively answering support tickets, redlining contracts or editing podcasts to teaching an agent once how to automate them. Inside enterprises he sees people moving toward “creating eval skills files” and training agents. Mercor's bet is to lead and build the technology for that shift, which is why he believes it can become a category-defining company.

    ▶ 1:02
  2. Mercor is not a marketplace: it sells RL environments, and experts are just one ingredient

    Brendan calls the biggest misconception that Mercor is an Uber-style marketplace taking a fee. What it delivers to labs is RL environments with three parts: apps (clones of Google Workspace and similar), worlds (the messages, docs and sheets inside them), and tasks (prompts plus rubrics, like a professor grading an essay). Experts populate worlds and tasks and are one moat, supported by a talent platform, a data platform called Studio, and an AI project manager.

    ▶ 5:38
  3. Agents made experts 3 to 5x more productive, but Jevons paradox expanded demand instead of shrinking it

    Brendan says an hour of an expert's time with an AI co-pilot now yields perhaps 3 to 5x the model improvement. Rather than reducing need for humans, demand grew exponentially, because “99% of knowledge still lives in people's heads”. Making teaching more efficient means more teaching across the economy.

    ▶ 7:41
  4. Human-taught data stays durable because models need humans to measure the frontier where they fail

    Brendan argues the market's durability depends on superintelligence timelines. As long as humans can do things models cannot, such as a persuasive meeting, a wet-lab experiment or fixing a car, models need human feedback to find their mistakes. He compares it to a student forced to grade their own essays. Math is an exception, since calculators let you check your work.

    ▶ 9:12

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.