Comments like these make me wonder whether we live in the same worlds.

I'm a cursor user e.g. and Tab completion is by far the most powerful auto complete I've ever used.

There's scenarios where you can do some major refactors by simply asking (extract this table to its own component while using best react practices to avoid double renders) and it does so istantly. Refactor the tests for it? Again. Semi instant. Meanwhile monocole wielding "senior" is proudly copy pasting, creating files and fixing indentation as I move to the next task.

I don't expect LLMs to do the hard work, but to speed me up.

And people ignoring LLMs are simply slower and less productive.

It's a speed multiplier right now, not a substitute.

If you complain that you don't like the code, you're not understanding the tools nor you can use them, end of story.

Sounds like you’re working in react. The models are orders of magnitude worse at less common frameworks and/or languages. React is so boilerplate-heavy that it’s almost a perfect match for LLMs. That said, every large react codebase I’ve worked in inevitably became really complex and had a ton of little quirks and tweaks to deal with unexpected re-renders and such. I avoid the FE when I can these days, but I’d be fairly surprised if LLMs generated anything other than typical performance-footgun-heavy react

Not in my experience, claude seems great at writing idiomatic and performant react. You can even highlight code and ask it to spot performance issues and it will tell you where you can improve rendering efficiency etc.

That's interesting. I haven't tried it, and my day job moved on from React some time ago. Most of what I do these days is Rust, which all the models I've tried are pretty miserable at.