If we were using AI as a tool, I would 100% agree with Eric. But we’re not. We’re pretending it’s a precision C&C machine when it’s the most Rube Goldberg way to produce sloppy work in existence.
The edges are sloppy. The cuts are sloppy. The precision is all over the place. And
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem
why are we not talking more about how agents make it easier to read as opposed to write code? you can literally now scan your entire repo and ask it to describe the entire logic path of a single route, how different configs fit together, find the PR where feature X was introduced
"You already ship code you don't read" is a bad argument because in the past _somebody_ read it, or at least they did until recently. That's what the OSS supply chain was always about - maintainers had to *earn* trust.
