I find myself using LLMs mostly to jump start into working and "saving" the energy that I need to start working on a new task. Whether the LLM produces something good or not, I am now in the middle of work.

It works best if I'm not already exhausted because a that point LLMs output just confuse me with a lot of text and code.