Will Brown on Scaling Agent-Written Code

Will Brown, Research Lead at Prime Intellect, posted on X that massively scaling agent-written code remains feasible and likely optimal when the code stays fully human legible. Humans could then validate intended behavior through quick spot checks. The message included an attached screenshot showing the lead paragraph of the English Wikipedia article on the PCP theorem. Brown works on agentic reinforcement learning and open-source AI infrastructure including the Verifiers library. The evidence contains only this post and its attachment as his direct statement on the topic.