I feel like i want a more intuitive, natural process. Purely for illustration -- because i have no idea what the ideal workflow is -- I'd want something that could allow for large autocomplete without changing much. Maybe a process by which i write a function, args, docstring on the func and then as i write the body autocomplete becomes multiline and very good.
Something like this could be an extension of the normal autocomplete that most of us know and love. A lack of talking to an AI, and more about just tweaking how you write code to be very metadata rich so AIs have a rich understanding of intent.
I know there are LLM LSPs which sort of do this. They can make shorter autocompletes that are logical to what you're typing, but i think i'm talking about something larger than that.
So yea.. i don't know, but i just know i have hated talking to the LLM. Usually it felt like "get out of the way, i can do it faster" sort of thing. I want something to improve how we write code, not an intern that we manage. If that makes sense.