What's the one thing you never let AI do? My rule for drawing the line.

I've been building with AI for a while now, and I've noticed something: the projects that work best aren't the ones where I let AI do the most. They're the ones where I've done the hard thinking first.

Here's what I mean. Last week I was mapping out a feature set, and I realized I'd spent two hours prompting Claude to help me decide what to build. It was... fine. But it felt hollow. When I stepped back, predefined exactly what problem we were solving and what constraints mattered, then asked AI to help execute against that? Completely different quality.

So I've settled into this rule: predefine everything you can. The problem statement. The constraints. The principles. The success criteria. Then let AI do the few things it's actually better at — rapid iteration, spotting patterns you missed, exploring variations fast.

But there's always something I keep for myself. The line in the sand.

For me, it's the "why." I won't let AI define why something matters or who it's for. That clarity has to come from real judgment, from understanding your users or your own convictions. Everything else can be negotiable.

What about you? Where do you draw that line? Is there something you've decided you're never handing to AI, no matter how good it gets?