Using AI tools is the same as micro-managing junior devs. You spend a shitton of time to get the required result. And to do this better and faster, you need to break the task into distinct simple subtasks. Otherwise, you are fucked.

Considering this, it's arguably better than doing stuff yourself. I guess it depends on whether you are going to apply the obtained knowledge in the near future. Most likely I won't be doing geometry related stuff after I finish with my splash page update, so it's fine. But otherwise it's better to read the docs and do the dirty work.

By the way, I mean working with something unknown to you.

Ah, and not to forget to prompt the same question several times. It's completely normal that it can give a better response without you giving more context. This is actually a bit annoying. One would think that the more data you provide, the more precise the response should be. But it's absolutely not the case with these LLMs.