I’ve typically used LLMs in non-editor environments in an advisory capacity - there is a lot of back and forth but very little planning. After my latest project got too complex, I had Claude do some true planning. After some relatively short back and forth about refactoring everything, it came up with almost 900 lines of our ‘new plan’.
I’m not about half way through implementation and I’ve basically just been clicking accept. For projects where you’re going through a large foundational build with little risk of messing up something that doesn’t exist - i’d love to be able to just let it run and build.