How this site was built
If your work isn't written down somewhere a machine can read it, it might as well not have happened. That's the premise of this site. More and more, the first thing to read about you is a model summarising you for someone who hasn't met you, out of whatever it can find.
Anything I did under an agreement stays off the record, and that rules out a good part of the last ten years. Everything else deserves a small write-up, and nobody's got time for that. So I threw an LLM at it.
Some context, because I've had a varied career and the rest of this page assumes you know it. I spent most of my twenties as a designer running a small web agency in Birmingham, 3form, and I ran a record label, Type, on the side; the catalogue's here. Then I co-founded Makeshift, a London venture studio that built fifty prototypes in two years, one of which grew into Product Hunt.
Along the way I've made data art, given talks and written a few hundred posts. Since then I've led product and engineering at On Deck and built the AI tools at Oak National Academy, and now I'm building Arcla and helping at Assembly.
By 2023 this site had shrunk to a single page with a few client logos on it, and everything else had gone. Not lost exactly, but scattered: old copies in the Wayback Machine, posts on Medium, code on GitHub, a Dropbox account and a few drives in a drawer. In the summer of 2026 I put it back together from all of that, with AI agents doing the reading and me doing the checking. This page is what happened.
Why I bothered
Most of what I built in those years spoke for itself in the room where it happened, and it does not speak for itself any more. I think that is a strange thing to have to accept. I'd rather write it down than argue with it. So the site is built for two readers: a person who wants to know what I've done, and a model doing the same job on someone else's behalf. There's a plain text copy of the whole record for the second one.
Where it all came from
The Wayback Machine had copies of most of my old sites, right back to the first one, and GitHub had the code. Medium and my old blogs had the writing, Companies House had the facts about the companies, and a Dropbox account and a few old drives had the pictures.
Nothing on this site was written from memory: if a fact is here, it was read from a page somewhere, and the entry says which page.
What the AI did, and what I did
Agents read all of it, roughly one per project, and wrote each thing up as an entry: a company, a job, a project, a record, a talk, a piece of writing. They worked to one rule, which is that if the page doesn't say it, the entry doesn't say it: no tidying, no guessing, no filling in the gaps. Where two pages disagreed nobody picked a winner, and the question came to me instead.
Then I read it. I've corrected names and dates, cut a few things the agents got wrong, and cut a few things that were never mine to claim - a theatre company's tour dates once turned up under my name because I'd built their website. I'm still going through it, and some entries are still waiting on me.
The numbers on this site are counted from the entries each time a page is built, never typed in. Today it holds 687 things made between 1995 and 2026, read from 115 old sites, with 1,420 links back to the pages they came from, and 63 pieces of coverage. If a number is on the page, you can check it.
How to read it
Some things are named but not linked. If all the record holds is a name and a date, there's no page worth sending you to, so the name sits in plain text. The page still exists if you have its address, but it doesn't invite a click that goes nowhere.
Now offline. The thing has gone from the web, and where the Wayback Machine kept a copy the entry links to that instead and calls it the archived copy.
Some dates are hedged. 'By 2012' means a page proved the thing existed by then, not that it was made then, and 'c. 1997' means I remember it but can't prove it. The hedge stays on the page so a guess never passes as a fact.
What's still missing
Quite a lot. Pictures, for a good many of the projects, because the old sites are gone and the archive kept the words but not the images. The years before the web kept records. Anything I did under an agreement, which I've left out rather than hidden. And the entries that are waiting on a question only I can answer, which stay off the main pages until I've answered it.
So if you knew me in one of those years and something here is wrong or missing, I'd like to hear about it.
Do it yourself
The whole procedure is written up as one file you can hand to a coding agent and point at your own history. It says where to look and in what order, what an entry looks like, and the rules that stop the agent making things up, with my name taken out.
archive-archaeology.skill.md Plain text, a little over 20 KB. Open it in the browser or save it.
Licensed CC BY 4.0: use it, change it, credit the source.
If you build one, I'd like to see it: hi@stef.io