Depends a ton on what for. For small applications, scripts, tests, self-contained things, or repetitive tasks (think refactoring), it’s basically 100% AI+review (often tests you have to tweak and carefully check cause it’ll write tests that pass but don’t actually test anything with it being obvious at all!). For high-level tasks in large codebases, especially starting to design a new API or feature, it’s somewhere between 0 (doing it yourself to explore an approach while being more directly exposed to the changes) or using them to explore but then rewriting chunks after you’ve explored different options.