The Conductor

Every Hat in the Room is about one person wearing every role. This is the next shape: you stop wearing the hats and start handing them out.

Alex saw it happen on a Tuesday afternoon. He was building a landing page for a side project — three AI agents working the same folder. One handled the copy. Another built the components. A third reviewed the output and flagged inconsistencies. They didn't talk to each other. They talked to the folder.

At one point, the copy agent generated a headline that contradicted the feature list the component agent was building from. Alex caught it. Not because he understood the code — he's never written JavaScript in his life — but because he knew what the page was for, and the headline didn't match. He corrected the copy agent, re-ran the component agent, and the page snapped into alignment. That was the wrong note, and he heard it because he was the only one listening to the whole piece.

His reaction wasn't "that's useful." It was something closer to awe. He'd already experienced wearing every hat himself, and that had felt like discovering a new dimension. This was different. This was: the hats can be worn by different agents, simultaneously, and you just... conduct.

The shape is orchestration. A conductor doesn't play the violin or the timpani. A conductor holds the vision — the tempo, the dynamics, the shape of the piece — and each musician plays their part. The conductor's value isn't in any single instrument. It's in the fact that someone has to hear the whole thing at once.

That's what multi-agent workflows look like in practice. You set the vision. You assign the work. You integrate the results. Each agent is a specialist — one is better at research, one at code, one at prose, one at visual design — and the folder is the shared workspace where their outputs accumulate. You're the only one who knows what the project is for. The agents know how to do their part. You know which parts matter.

This is different from Every Hat because you're not switching roles anymore. You're delegating them. The cognitive load drops from "be the PM, then be the engineer, then be the designer" to "tell the PM-agent what you need, tell the engineer-agent what to build, tell the designer-agent what it should look like." Your job becomes judgment and integration — which parts to keep, which to throw away, which agent's output to feed into another agent's input.

The risk is the same as any delegation: you have to know enough to evaluate the output. A conductor who can't hear a wrong note is useless. A person orchestrating three AI agents who can't tell good code from bad code, or a strong argument from a weak one, is just generating confident garbage at three times the speed. The skill isn't in running the agents. It's in knowing what good looks like.

But here's what hit Alex: you don't have to be an expert in everything. You have to be an expert in your project. You know the vision. You know what the output is supposed to do. You know when the code doesn't match the intent, even if you can't write the code yourself. The conductor doesn't need to play the oboe. The conductor needs to know when the oboe comes in wrong.

That's the upgrade from Every Hat. First you learn to wear them all. Then you learn to hand them out and listen for the wrong notes.