Most AI pilots never make it to production. I build the ones that do.

I want to set out plainly why this site exists and what I am trying to do here.

We are living through something the human record has no precedent for. Systems are being built from the accumulated written knowledge of our species, and we are now working next to them. Whether that constitutes a new kind of entity or simply a very capable tool is a question I am not qualified to settle, and I notice that people I respect disagree about it. What I am certain of is that the relationship does not run in one direction only. In learning to build better tools, we are being asked to change as well.

Some of that change is already visible. Work is being reorganised, and some of it is disappearing. The energy and water demands of large-scale compute are real and are being paid for somewhere. Less visible, and to me more interesting, is what is happening to how we think. When a system will produce a plausible answer on request, the habit of sitting with a problem long enough to understand it becomes optional. I do not know whether our cognitive capability is being extended or eroded. My suspicion is that it depends on how each of us chooses to use these systems, and that most of us have not yet made that choice deliberately.

I am not neutral about where this goes. I believe organisations can adopt this technology safely and ethically, and be better for having done so. That belief is a commitment rather than a prediction. It requires keeping people at the centre of the decisions, particularly the decisions that become easier to make when nobody has to look at the consequences.

There is a line from Sun Tzu that has been quoted often enough to have lost some of its edge: “If you know the enemy and know yourself, you need not fear the result of a hundred battles.” I would keep the second half of it. The technology is not the enemy. The thing worth knowing is ourselves: our incentives, our blind spots, and how readily we hand over judgement when something makes it convenient to do so.

So the purpose of this site is straightforward. I go and find things out, I test them, and I write down what happened, including the parts that did not work. If that saves you the same detour, it has done its job.

I would also like to hear from you. If you have run into something instructive, whether it succeeded or failed, send it to me. Where I use what you send, I will attribute it to you by name and represent your account as you gave it to me. What I am after is a shared record rather than a broadcast, and that only works if it is built from more than one person's experience.