That's only if you don't preserve the results explicitly. If you're trying to delve into some new code without enough docs, I could imagine learning lots about the system along the LLM and then leaving that as documentation and/or agent files in the repo.