technology

I'm still under the impression that it's totally useless. I've written much better and actually more code since I stopped using LLMs (I had co-pilot for a bit).

I really do think that people have been fooling themselves with AI slop. It's really easy to feel like you're doing a lot when it's just constantly spitting stuff out on the screen.

I'm recently going back and re-writing a package that I used AI to help write like a year and a half ago. This is the one that made me abandon AI, since it just continued to fuck up patterns in ways that were incredibly difficult to fix. Then the package suddenly got a decent amount of users and I was locked in.

I came up with the high level concept and used AI to help scaffold and boilerplate out models and stuff from a spec, which should have been a sign that I was doing something wrong.

Now I'm rewriting the whole thing from scratch, and one of the big things I've noticed is that the types of things I'd use AI for before, I'm now writing build scripts for. Which makes maintaining the library like 10x easier. It also forces me to distil the library backend down into the simplest possible form to keep the buildscripts simple.

I'm also forced to be way stricter with my typing solution (Python) since I can't just offload "make this right" to the AI and actually need to use strict static checks to keep the layers properly synced. AI Python tends to be awful when it comes to type hinting if it even does it.