I was scared of this initially, but I've found that I mostly just use it for tedious code that I would normally procrastinate just because of how mind numbing it was (like adding another CRUD endpoint) or making a sweeping change across code that wasn't as simple as a find and replace.

For things that keep me interested, I just won't use LLM features. Sometimes at the end I'll have it audit my code and sometimes it'll catch something that can be improved.

Also test cases. It's not perfect, but a large chunk of that being automated there is very nice.