> We are in fact seeing software improvement, in the form of fewer bugs and better performance.

I would not say that software on average is getting less buggy and more performant. I think the only big exception is being able to bruteforce-find security vulnerabilities and similar defects, which LLMs are well-suited to. But when you look at the remaining 95%+ of the software market, where everyone is seemingly all in on AI, all I see is software becoming more janky and less stable. Everyone just ignores it because it's faster for now.

> We are in fact seeing software improvement, in the form of more and better features in existing products.

Maybe 'features' in terms of measure-turned-target metrics that management at software companies loves to use, but overall I'm not seeing an explosion of features that's inconsistent with the rate of improvement until today. And what new is here is more janky than ever before.

> We are in fact seeing software improvement, but the "we" is very limited.

Surely the people on a tech news website would know about this very useful AI-only software, even if they didn't start using it. It's funny how no one can even give some names.

> Many businesses' software-development processes have subsystems, especially political subsystems, that have not been accelerated with AI.

The entire industry is obsessed with AI, almost everyone is stripping out everything that's not AI and replacing it with AI. Doesn't seem plausible.

> Software teams are capturing a lot of the speedup in the form of more leisure or quasi-leisure time.

How many people did you talk to recently that said they're spending less time working?

> AI improvements in software sometimes push us to use and make software less.

In what ways? AI, by allowing you not to understand the solution, can carry a doomed project much further than a regular programmer would before realizing their mistake. If instructed to, the AI will just build and build and build until the thing implodes under its own weight. It gives you less awareness of what you're doing.

All of these are unconvincing. I think only one of the points has merit:

> We're making more software instead of making software better.

I think people try to avoid acknowledging that AI may be much better at improving the speed side of the equation than the quality side. Look around - all you see is demands for more things faster, even if it's total garbage in the end. They didn't want to bother with quality already, now they found an excuse to try to forget about it altogether. If it's wrong, we'll just regenerate the same garbage 50 more times, they say. Focus on shipping something now, they say.