Programming is still fun, at least if you are solving real problems and building real products.

The part of programming that sucks the most is untangling ambiguity. In my current app I had an ultra simple proxy convention, but it sucks. I have since settled on a different single convention to cover 6 different scenarios of traffic redirection, but it was a challenge to distill simplicity from chaos of the wilds and write all the various tests. It’s moving forward faster now but it was like starting over to get this rolling.

What I find is fun is building things that solve real problems I have. Then my problem is solved, I am paying nothing for it, and I can extend it however I want for any reason without downstream restrictions.

There is so much AI still cannot do in software. If you ask AI to build a faster WebSocket solution, for example, you will get more of what’s already there. I have a WebSocket library I wrote myself 4 or 5 years ago. I thought it was fast, turns out it wasn’t. I wrote my own perf tool for it and had a defect in the perf tool that spit out wrong results, but that defect showed me where the performance bottleneck was. Now my library is 40x-100x faster than other WebSocket libraries. I did use Claude to analyze my earlier perf bug, and it made the same mistakes I was making. If you want to go faster than popular battle tested solutions you have to be original and AI just isn’t quite there.