[jezek2]

I don't think it's wise to use it for anything. Even when chatting with an LLM you would have to check everything yourself which nobody would truly do. The generated code can't be really trusted (and nobody will review everything, quite the opposite). It can also have copyright issues.

People are allergic to articles and documentation generated/processed by LLM.

You're switching from an active role to a passive one, meaning your skill will suffer over the time. There is a huge difference between doing the things and thinking you know what it's doing. It's harder to review bad generated code because how polished it looks, compared to code made by humans where the difference is much more obvious.

Code assistants seem to work great when dealing with boilerplate, but wouldn't be better to get rid of the need for the boilerplate in the first place?