What I kept seeing
For several years I sat in due diligence rooms, reading the accounts and the products of software companies. It is a good place to observe from. You see how a company describes itself to the outside, and you see how it actually runs.
Before that, my first real job was to take a product into a market I knew nothing about and build demand for it from zero. A market does not listen to your plan. It only shows you whether people use the thing every day. That was the first time I understood the distance between a real need and a good story, and it is the lens I brought into every one of those rooms.
The step without a manual
Inside almost every company I looked at, the picture was the same. Most of the work had been turned into software long ago. Orders, invoices, scheduling, reporting, all of it ran on systems nobody thought about any more. And then there was one step that did not. It was done by the person who had been there the longest, by eye, from experience. There was no manual for it, because nobody had ever managed to write one. That step set the pace of the whole operation, and it was usually the most expensive thing in the building.
After late 2022, almost every company's materials mentioned AI. Very few had numbers that changed because of it. Where the numbers did change, it was almost never in a new chat window. It was in that one step, where an experienced person's judgement had quietly become cheaper, and the same person was now getting through far more of it.
That left me with a view I still hold. Businesses are willing to pay for new technology, far more willingly and far more steadily than individuals are, on one condition: the technology has to bring a visible gain in productivity. The reason is plain. A business has accounts to balance. Skilled people are harder to find every year and cost more every year. These are not trends to a business owner. They are the monthly bills.
The word visible matters. If everyone in the company gets a little faster, nobody pays for that. If the one step that used to take three experienced people now takes one, the arithmetic is obvious and nobody needs convincing.
For thirty years software could not touch that step, because software needs rules, and this kind of work has none that anyone can write down. What changed after late 2022 is not that machines became clever in general. It is that a model can now learn a judgement from examples of it, the way an apprentice does, only faster. For the first time, the step without a manual is within reach. That is the whole reason this company exists.
Why not the obvious place
AI is the most crowded place in the world right now, and most of the crowd stands in one spot: the model layer and the tools around it. I have no edge there and no wish to pretend otherwise. The other end is nearly empty. Taking a model into one trade, sitting with the people who do the work for months, and learning which step costs the most, is slow and unglamorous, and it does not look like a company that will be large next year. That is exactly why it is where I want to be.
So the rules of this company are few, and they are written down. We work on a problem from the first interview until someone pays for the result. We choose what to build by a short list of tests that live on paper, not in the mood of the day, and most of our own ideas are supposed to fail them. And I want what we build to still be here in ten or twenty years, still doing something real for the people who do that work.
Who I am looking for
We have picked our first trade, and it is dentistry. I am looking for three kinds of people. Dentists and laboratory technicians who make expert judgements all day and can tell me exactly which step costs the most. Partners with a product, a channel or a market who want to see one thing through. And early-stage founders building for work like this, wherever it is.
If that is you, write to me. I read every note myself and reply within a week. If there is a fit, the next step is a conversation, not a process. If there is not, I will say so.