My problem is that, for a surprising number of applications, it's taken me longer to have the conversation with chatgpt to get the code I want than just doing it myself.
Copilot and the likes are legit for boilerplate, some test code, and posix/power shell scripting. Anything that's very common it's great.
Anything novel though and it suffers. Did AWS just release some new functionality and only like 4 people have touched it so far on GitHub? Are you getting source docs incomplete or spread out amongst multiple pages with some implicit/betwen-the-lines spec? Eh, good luck, you're probably better off just reading the docs yourself or guess and checking.
Same goes for versioning, sometimes it'll fall back into an older version of the system (ex Kafka with kraft vs zookeeper)
Personally, the best general use case of LLMs for me is focus. I know how to break down a task, but sometimes I have an issue staying focused on doing it and having a reasonably competent partner to rubber duck with is super useful. It helps that the chat log then becomes an easy artifact to more or less copy paste, and chatgpt doesn't do a terrible job reformatting either. Like for 90% of the stuff it's easier than using vim commands.