I’ve found that when I ask AI to do something for me that I know how to do myself - but would rather not spend the time doing - there is a not insignificant chance that the AI will return a subpar result, which I can usually tell rather quickly. Either by glancing at the code, or trying to compile it and getting an error.

This happens frequently enough that it creates a real disincentive for me to use AI for anything that I already know how to do - and use it exclusively for things I don’t know how to do.

It’s deeply frustrating to realize you just wasted 20 minutes posting error messages into Claude when you could’ve just locked in and written it yourself.