What AI Agents Still Can't Do (The Last 10% Is Still Yours)
The promise of AI building tools keeps inflating: “describe anything, it builds anything.” Every few months someone ships a demo that looks like the software writes itself. Here’s the honest counterweight, from actually shipping with these tools: agents are superhuman at volume and untrustworthy at judgment. They can generate more code, faster, than any team you’ve seen. They cannot decide what’s worth building, what’s actually correct, or what’s good. That last stretch — call it the last 10% — is where the product is won or lost, and it’s still entirely yours. ...