Slow is a strategy

A model writes a working function in the time it takes to read this sentence. It drafts the memo, the migration plan, the pitch. Speed used to be a thing you could sell. Now it comes for the price of electricity, and a capability that cheap is not an advantage. It is the floor everyone already stands on.

So I stopped competing on it.

The name I work under is homo faber, the one who makes. Not the one who ships first. A person at a bench, turning a real object in real hands, is the oldest picture of good work we have, and it outlasts the arrival of machines that type faster than we think. The bench does not care how quickly you produced the wrong thing. It cares whether the thing holds.

What stays scarce is the judgment about what to build. A generated answer is only as valuable as the question underneath it, and the question is almost never the one you were handed. So the first thing I do is descend: two weeks inside the client’s problem, in the systems and the margins and the code and the people who run it, until the real question surfaces. It usually contradicts the brief. That fortnight cannot be compressed, because it is not producing output. It is producing the ability to tell a good instruction from a plausible one, and no model will do that for you, because a model answers whatever you ask with equal confidence.

This is what deliberate slowness protects, and it is not caution. Caution drags on everything alike. This kind is selective: it spends time only where judgment gets made, and moves fast everywhere else.

The first company I built with another doctor, AIDA, tried to teach a machine when to release a unit of red cells. The brief looked like forecasting: predict how much blood each hospital would need, and keep it from expiring on the shelf. I spent the slow weeks at the actual decision instead, beside the people making transfusion calls, and the real question sat underneath the one I had been handed. Wastage did not come from bad forecasts. It came from units transfused that never needed to be. The scarce thing was not a sharper prediction of demand; it was a better answer to a bedside question the brief had not thought to ask. None of that came from speed. It came from two doctors sitting with a decision until they understood what it really was.

That discipline followed me from the blood bank to the energy exchange, where the hard call at Respect Energy was restraint: knowing what must never leave the building at all.

So I take few clients, I go unreasonably deep, and I stay until the thing runs in production under a real load, not until it demos.

A model generates in a second. Knowing it was the right thing to generate takes months, and no one has found a way to shorten them, only to skip them and pay for it later. Everyone is racing the machines down to the floor, where every answer is free and not one of them is yet the right one. I work above the floor, where a good instruction still has to be told from a plausible one, and that telling has stayed slow. Slow is not what the work costs. It is the work.