Yes, thank you! This is exactly my experience. It requires immense effort and concentration, maybe more than before AI, where
- the next ideal step progressively manifested itself to me during coding
- I actually had perfect knowledge of the codebase and didn't have to work as hard to reconstruct it in my head each time
- coding itself is half a mechanical job. AI does the mechanical job itself well, leaving you only the hard thinking.
- AI prose isn't easy to parse, at least to ne.
- Prompting AI until it gets the optimal solution / implementation takes time and revisions.
Something like that.
I keep thinking that it's an issue of my approach, though. I've been trying to improve my AI workflow as much as possible for months, every day, but still, I'm not content with it. I might have gotten stuck on a local optimum. Maybe I'm writing prompts needlessly detailed. Maybe I should read more of the code, because I don't, really, making me not know the actual codebase and how things work.
It's not easy, this whole thing. And honestly, sometimes I wonder - after accounting for the total time, including revisions, mental reconstructions etc - if it wouldn't have been done faster by me in the pre-AI world.