At least at my company (~100 engineers), we're trialing copilot and have steered away from using GPT at all over legal and compliance concerns. Haven't really spoken to anyone who's made it part of their core workflow. Most of what I've heard outside the company is that the stuff is very useful for senior engineers but if you don't know what you're doing it can lead to some pretty embarrassing issues.
I would say it makes me roughly 4x as productive as a software engineer. It's all about knowing how to use it. I do agree with the sibling comment that it helps a lot if you are already senior since you can immediately tell what is wrong/garbage that it is outputting.
I wonder what kind of stuff you're working on, that this makes you 4x more productive. For my work (stream processing, analytics, in Clojure), it's useless, AFAICT.
It's only really worthwhile on mainstream coding patterns and paradigms. If you're in a niche then generally the results tend to be very hit and miss.
This is not that surprising. There's just not enough source material for it to have learnt from and generalise.
I have a friend who does embedding programming in C++ - and he finds it occasionally useful but mostly it fails. He has the double whammy of being in a small niche and a lot of embedded people still writing in C (and doing it in very odd ways).
I would mostly agree for this, if you need it for some general patterns it's really good for that. Things like "how do I implement Singleton properly in Python contextually in this codebase". I can't use it as much on things post 2021, though, for instance.
I think in general it's just really good at Python, so it's a boon if you code in Python. I was using it for Kotlin the other day and while it helps, it's not as immediately good in my experience.