Yesterday an AI agent built a complete feature for Phora in 12 minutes. Routing, validation, UI — all in place. I could have deployed it immediately.
I didn't. Because the feature wasn't needed.
Good code. Good structure. Wrong decision. Users hadn't asked for it — I just thought it might be useful. The AI executed what I told it to, fast. The problem was what I told it.
Speed is a multiplier
If you know what you want to build, AI multiplies your speed. If you don't — it multiplies your steps in the wrong direction. You get to the wrong place faster.
I see this daily. AI agents can assemble entire modules from a well-crafted prompt. But behind the prompt there needs to be a decision: why this, for whom, and what's the goal. The AI doesn't formulate that. I do.
What can't be delegated
There are things AI is perfect for. Boilerplate, tests, refactoring, formatting — machine work for machines. I've outsourced all of it. I'm faster because of it, and I waste less energy on things that don't require thinking.
But there are three areas where I don't ask:
Product decisions — why this feature and not the other. Design taste — what's good, not just what works. Business context — what matters to this specific client, in this specific situation.
These come from experience, context, and taste. They can't be learned from a prompt.
I decide
I'm not against AI. I rely on it daily. But I know exactly where the line is: it executes what I say — faster than anyone else. But what I say is on me.
I decide. The AI executes.
There's always a next level.
If you like what you see — whether you're building a product or a team — I'd love to hear about it.