A very subtle but important bug in open-source lattice Boltzmann code. Claude didn't give up until it found it. Later they realised they could have just had Claude write a whole new solver.
Result Back when Claude was still cheap to run
“I thought mine was when claude found a very subtle but important bug in some open source LBM code I was using. It ground at it for hours and didn't give up until it found it.”
Spotted: found in the wild. Details are as reported by the source above and have not been independently verified.
They were using some open source LBM (lattice Boltzmann method) simulation code and set Claude on a subtle bug in it.
Claude ground at the problem for hours and didn’t give up until it found the bug, back when Claude was still cheap to run. They thought of it as their big moment with AI, until a conference showed them they could have simply asked Claude to write a whole new LBM solver instead.
Point Claude at a subtle, hard-to-reproduce bug in an existing codebase, ideally one that's been resisting your own debugging, and ask it to keep investigating until it finds the root cause rather than stopping at the first plausible explanation. Let it run for as long as it takes rather than cutting it off after one or two attempts.
A starting prompt written by Agents in the Wild. Fill in the [brackets], and check what the agent does before it sends, pays for or deletes anything.