I regularly find the code output of opus and gpt 5.5 to be garbage. Overly verbose, unnecessary abstractions, strange duplication of concepts across objects, unnecessary copying of objects and creation of objects. I have found its much more useful to just ping pong some ideas, have it generate helper methods, and do the code implementation by hand.

I guess I am a stoneager.

not to throw this word around, but,

> Only stoneagers would say that they are better than a good AI.

projection? lack of confidence in your own abilities? why make such a sweeping statement

This thread is in the context of community PRs in open source projects. So it's not about AI or not, it's about maintainers using AI vs random contributors using AI.

My point is that with AI, where the actual code generation is easy, there's little value in community PR contributions anymore.

> Only stoneagers would say that they are better than a good AI.

I am only a bit above average and I clearly still write better code than a good AI.

The only question left in my mind, alas, is whether that is enough to earn a living.

I mean: it is clear that in every domain except for programming, a talented XYZer can do better than an appropriate LLM trained to do XYZ (except perhaps in some absolutely exhausting pattern recognition tasks).

So I am not sure why we see our own field as different. A sort of inverted Gell-Mann amnesia?