Ask HN: Am I using AI wrong for code?

I have a nagging feeling that I'm missing out on productivity gains by under utilizing AI for code. I've used copilot and a few other tools but for the most part they didn't click with my workflow, or ended up leading me down paths that wasted time.

The only thing I do now is chat with Claude via the web chat to brainstorm architecture, tool recommendations, and to get small well defined code snippets or boilerplate to copy into my projects. Don't get me wrong, I think this has massively improved my productivity, and made me less scared of driving into projets with unfamiliar languages or stacks. My most useful interactions have been when I use my own experience to point out flaws with the AIs suggestions and ask questions. Its helpful but I feel like I'm missing out.

Given how powerful LLMs are plus options like RAGs and fine tuning, I feel like they should be more capable of accomplishing large chunks of work with supervision. I've certainly seen reports of entire reafactors, features, or language migrations being done with the help of AI.

My question is: what tools are people using?

What I have in my mind is a tool that can suggest and, crutially, iterate on edits to multiple files at once, include context based on chosen docs or codebases, and be able to branch conversations so that I can clarify and learn without derailing the main thread. I started working on a CLI that can do this but I'd rather use something off the shelf.