Comment by djx22

Actually, this paper shows that the LLMs create a lot more slop and degradation of the code after the first prompt. https://arxiv.org/html/2603.24755v1.

So iterating with the LLM will simply make it to produce worse and worse code.