Ask HN: How are you using AI / LLMs for coding?

I keep going back to RooCode because good ergonomics with VSCode and personal familiarity.

Recently tried "agentic" coding with Claude Code Max after being pestered by a colleague to give it a try. I had no luck with it. Letting it loose on a project led to many disastrous commits. A decent change was frequently followed by some disastrous dog's breakfast of a change where it tried to rewrite most of the code trying to fix some minor problem.

However, restraining it and then having to babysit every change made it no more effective than RooCode albeit with worse UI. In the end I gave up and cancelled the Claude Max subscription.

I do not believe people claiming they can make it work in an autonomous fashion. Even with a good scaffold of tests it lies about compliance or disables tests that don't pass.

I have had success with Cursor the most. I don't quite like the interface for Claude because it is terminal based, which is kind of poor UX IMO. But I still use it here and there.