I am not a founder. I started anyway.

For most of my working life I was the person who had ideas on the drive home and did nothing about them. Not out of laziness. The gap between I think this is a real problem and I have something a customer can use was measured in people I did not have and money I was not going to raise.

That gap is what closed. Not the ideas — I have no more of those than before, and mine are no better than anyone else’s. What closed is the distance between having one and finding out whether it holds.

It started with my own admin

I run a small business. Like everyone who does, I lost the same hour every week to the same things: an inbox that needed sorting before it needed answering, documents whose numbers had to be retyped somewhere else, follow-ups that lived nowhere except my memory and therefore sometimes did not happen at all.

So I rebuilt that. Not as a project — as self-defence. Email that arrives already triaged. Documents in, structured data out. Checks that run on a schedule whether or not I remember to trigger them.

It kept working after I stopped paying attention to it. That is the only test I trust.

What actually changed

I can now run, in a week, the kind of work that used to be a small team’s month: read a market properly, draft the argument, attack the argument, build the thing, put it in front of someone. Not because I got better. Because the volume — the reading, the drafting, the second and third pass — is no longer rate-limited by one set of hands.

The constraint moved. It used to be capacity. Now it is judgement. That sounds like a smaller problem. It is a much larger one, and nobody sells a subscription for it.

What did not change

Three things, and they are the three that decide whether any of this works.

• Knowing which problem is worth solving. A model will help you evaluate a problem you hand it. It will not tell you that the problem you are excited about is one nobody pays to fix.
• Being willing to hear no. The temptation is to keep asking until you get an answer you like. There is always another framing that makes the idea look better.
• Carrying the consequence. When a customer is let down, no part of the operating model feels that. I do.

The trap

Here is the thing I got wrong first, and I suspect most people do. AI makes it extremely cheap to build the wrong thing extremely fast. Early on I produced a great deal of finished-looking work that was not aimed at anything real. It looked like progress. It was motion.

So before building anything else I built the sequence — discover, verify, challenge, build, deliver, learn — with a bar between each stage that evidence has to clear. It is not sophisticated. Its entire job is to stop me shipping enthusiasm.

Why any of this is public

Partly because a bar you publish is harder to quietly lower. Mostly because the version of this story that usually gets told — the one with the origin myth and the inevitability — is not the useful version.

The useful version is that an ordinary person can now test a commercial idea properly, alone, in weeks. That is genuinely new. It is also not the same as an ordinary person being able to build a good company, and pretending otherwise helps nobody.

So this is the log. What I tried, what the evidence said, and what stopped.