AI tends to over-engineer more than I would. It’s kind of nice, because the code is often robust today, but it’s a lot to maintain tomorrow. I usually have to walk it back after a first pass, streamlining things, simplifying database changes. There’s a cost to doing too little and doing too much.
@manton I see it in the following way: AI is an adviser. It is more skilled than me and you, but definitely less experienced. You know your product better; I know mine. AI doesn’t know everything we know about our projects. We can hear and think. And make our own judgement on what will work today, and what we need to improve to make it work tomorrow. That’s it.
@manton I’ve just started to give github.com/DietrichG… a shot, as I find myself often getting in review loops fixing things that aren’t always clear they’re quite worth it.
Also trying to play with github.com/obra/supe… but not sure if it’s too heavy
