[layer8]

That isn’t necessarily true. LLMs could be good at finding issues that human miss when coding, while being bad at catching issues that they themselves miss while generating code.