In my bubble coders find LLMs least useful. After all we already have all kinds of fancy autocomplete that works deterministically and doesn't hallucinate - and still not everyone uses it.

When I use LLMs, I use it exactly as Google search on steroids. It's great for providing a summary on some unknown topic. It doesn't matter if it gets it wrong - the main value is in keywords and project names, and one can use the real Google search from there.

And it isn't expensive if you are using the free version