An AI analysis of my own practice · still forming
Twenty years of building things, and I've spent the last few weeks asking AI to make sense of all of it — what I sell, what I've built, where the evidence is thin. It read everything public about me and produced the analysis below. It is confident, it could only see what I'd published, and it has already been wrong thirteen times over — and it cannot see where I'm taking this next. That's why it's here rather than in a drawer. I'd like some human perspective before I act on it.
Before you push back — ten minutes of audio or thirteen of reading gets you the fuller picture this page can only summarize.
An AI's podcast-style take on the analysis below, not a recording of me — if you'd rather listen than read.
Capability evidence, the portfolio matrix, and every open question in full.
Download the brief PDFEvents & venues platform
Bookings & payments — a quick experiment
Productised services
Games — Incrementic Game Studio
Partnerships & standalone
Worth saying plainly: none of these were stupid mistakes. They were things I had never published. The analysis was as good as my public record, which turns out to be the actual finding.
Every version of this analysis reaches the same conclusion: everything terminates in one person, so cut before you grow. That has been true for twenty years. It is becoming less true.
I've started running real work through AI agents and coding tools, and the change is not incremental. As I write this there is a report being drafted, three products being coded and correspondence being composed — at the same time, by me. The next step is to go all in on it.
It helps that I've spent twenty-five years scripting and automating on Unix. Orchestrating agents is shell, filesystem, processes and version control — I can read what they did and fix it when they're wrong, which is where most people stall.
The goal is to spend less time as the person doing the work and more as the one orchestrating it. If that holds, the breadth this analysis reads as over-extension may turn out to be the point rather than the problem. If it doesn't hold, the analysis is right and I should be cutting. That is the single most consequential open question here, and it is the one an outside view would help most with.
An analysis built from public evidence is confident in exactly the places it should be least confident. It can see what I published; it cannot see what I earn, what I turned down, what a client actually said, what I've quietly stopped doing — or, as it turns out, whatever I shipped last month.
If you've worked with me, hired me, been taught by me, or just have a view — the useful thing isn't agreement. It's the correction.