@nikink So I feel like there's a problem with the way this kind of thing works.

A lot of experienced developers are assuming that to use AI they have to, or they're forced to, integrate it directly with their code editing and use it there.

And I'd generally argue that trying to make that skill that they're already really good at, slightly faster, doesn't work. Either it tends to give them suggestions which aren't really appropriate to the very specific and technical work they're doing, or they're (maybe unconsciously) fighting against it on things like coding style.

OTOH, as a developer with a reasonable amount of experience where I've found AI tools most help me is with answering questions that I'm not super familiar with. I'm learning about how to work with git submodules, or checking on whether Python dicts are async-safe. It would take me longer to find the right article, from the right search, and read through to find the right bit of information; with AI I can ask exactly what I need and get a customised, and friendly, response.

I think too many experienced, and 'hot-shot', developers are assuming that the thing they need to speed up most is doing what they're already familiar with, because that's where (they think) they spend most of their time. And I'd agree, in principle; I think we're a long way yet from coding assistants to be able to really 'read my mind' and know what I'd want to write there.

That's not where I think we actually spend a lot of our time though. I spend more time fighting with our build systems, or trying to learn how other things work, or just reading through my email. And many of those things still can't really be automated or left to AI.