Articles
The economy will become an RL environment machine
8 November 2026
Reinforcement learning in recent years has become immensely important to not only engineers in AI, but the broader economy, as it continues to drive further performance of LLM models as the seemingly last scaling step to AGI. One of the clearest public statements of that economic shift came from Brendan Foody on Lenny’s Podcast.
Foody is the CEO and co-founder of Mercor, an AI data company that organises domain experts (physicians, lawyers, engineers, and many other fields) to help frontier labs and enterprises train and evaluate models. In practice that means rubrics, expert judgment, and realistic RL environments: worlds where a model can act, get scored, and improve. Mercor sits between human expertise and the labs that need a measurable definition of success outside academic benchmarks.
On the episode Why experts writing AI evals is creating the fastest-growing companies in history, he puts the loop in one line:
“It’s highly likely that the entire economy will become an RL environment machine, building out all of these worlds and contexts.”
Watch the clip
The quote lands about fifty seconds in. The embed starts there; open the full episode if you want the surrounding discussion of evals, verifiers, and why labs pay for expert-built environments.
What he is pointing at
The usual AI story of the last few years is job displacement. Foody’s counter is a new category of work: people who do a valuable task once by encoding it as an environment and a reward, then let agents repeat it. Tax filing, a data-room analysis, a support ticket pattern. The scarce asset becomes the world definition and the measure of success, not another copy of the same manual workflow.
That is the same contract as any reinforcement learning environment: observation, legal actions, reward, and when a round ends. Academic games still matter for research. The economic claim is that professional work itself becomes the next batch of environments labs and companies need to build.
Foody later expanded the same idea in Mercor’s essay The Economy will Become an RL Environment Machine, and in conversation with Tyler Cowen. The Lenny episode is the public clip people usually share for the one-sentence version.
Why this matters if you build environments
If the economy is becoming an RL loop, someone still has to invent each environment: what the agent may see, what moves are legal, what counts as success, and how a person checks a rollout. That inventing step is separate from training a policy. Play the world first. Package it for a trainer later.
For a practical walkthrough of that contract, see How can I make a reinforcement learning environment?. For describing one in ordinary language and judging the draft by playing it, see How to vibe code a reinforcement learning environment.
Where Viscerl fits
Viscerl is for the inventing step: describe a custom interactive environment, get a playable draft in the browser, and expose observe, reset, and act over MCP. It does not replace Mercor’s expert marketplace, and it does not train a policy. Use it when you need a world you can play and call before packaging it for a lab or a trainer.
Questions
Who is Brendan Foody?
Brendan Foody is the CEO and co-founder of Mercor. On Lenny’s Podcast he argues that as labs saturate academic benchmarks, the scarce work becomes building real-world evals and RL environments, and that much of the economy may reorganise around that work.
What is Mercor?
Mercor is an AI data company that matches domain experts with frontier labs and enterprises to train and evaluate models: rubrics, human judgment, and RL environments grounded in professional work. Their tagline is organizing human intelligence to power the AI economy.
What does “the economy will become an RL environment machine” mean?
Foody’s claim is that knowledge work shifts from repeating a task forever to teaching an agent once: experts encode a workflow as an environment with a way to measure success, then agents can run that workflow many times. The bottleneck becomes building those worlds and reward signals, not only writing more model code.
Where can I watch the full conversation?
The episode is Why experts writing AI evals is creating the fastest-growing companies in history on Lenny’s Podcast, with Brendan Foody. The YouTube link is embedded in this article; the quote lands near the start of the conversation.