I usually try to iterate - read available documentation (e.g. comments in a config file, product documentation,...) and try to find stuff out. If I get stuck, an LLM answer may be confidently wrong, but it may give me some new pointers in which direction I should go next. Or maybe mention some buzzwords/techniques/concepts that I might need to investigate further.

As it's underlying concept is pattern recognition it might not be completely correct, but more often than not nudges me generally in the right direction. Bonus: Now I probably learned some things that will help me later on.

So far I never had something a little more complex that an LLM gave me a correct solution for. But as I like to tinker, explore and learn for myself, I'd probably hate getting a complete working solution without any work I did myself.