Fair. But the approach of leaning more towards "doing it yourself" than "having the AI do most of the work" boils down to how much you trust an AI. Like already I'll delegate work to another dev, because I trust that if I explain something clearly they'll come back with something good. What's holding back AI is that it is in some ways the best developer imaginable, in other ways an incompetent toddler. But with Claude 3.7 Sonnet I have felt myself making that leap to trusting AI for writing code. Granted I am trying to be ahead of the curve as I do so, I have AI-specific reasons for doing that. But eventually AIs will get better and developers will find that asking for snippets instead of features is a less effective use of their time. I also think not only will AI improve, but our architectures will be reshaped to better accommodate AI. Things absurd to ask a dev because you'd be wasting their time is just another 5 words you tack at the end of a request from an AI. In the short term, we'll be writing nano-services, not in the distributed systems sense, but just tiny modules to give AIs a fighting chance to give stuff a bit bigger than a snippet, and a bit smaller than what causes their contexts to get overwhelmed, at which point all bets are off. Personally I avoid overloading their contexts like the plague. Which leads me to my original question, because the only AI IDE I tried is Trae (because I'm weird and don't do the normal thing) and the idea that it was proactively feeding in some of my code to the AI, and summarizing my conversations as a favour to me, does not understand how precious I find not overloading AI's nano brain. I'd rather micromanage that activity myself than to trust an IDE to do it. But that's why I was asking, I'm seeing if I can find anyone who is really sold on Cursor or another being the best thing since sliced bread, so I can see how they work around these pain points. I can report so far I haven't met anyone who has gone all-in.