Building
I think best by making things. These are the three that taught me most.
Afterlife Carbon
I founded Afterlife Carbon to intercept end-of-life carbon-fibre sports equipment and turn it into premium consumer products with verified provenance — giving sports equipment the afterlife it deserves. Not a recycling company: a premium consumer brand that happens to solve a waste problem. Carbon fibre is a remarkable material with a dismal end of life; only around 2% is recycled globally.
The proof-of-concept product is (Re)Orbit, a line of sunglasses made from reclaimed carbon fibre: handmade by an eyewear manufacturer in Northern Italy in over a hundred production steps, Barberini polarised lenses, and an NFC tag in the frame that traces each pair to the equipment it came from. British recycling, Italian craft, provenance you can scan.
Building it took me into rooms with British Rowing, SailGP, Cervélo, Ribble Cycles and materials innovators like Fishy Filaments — and taught me more about positioning, provenance and brand than any book has.
The autonomous company builds
Twice in July 2026 I gave an AI agent a single instruction document and my full knowledge system, and asked it to build a complete company from scratch — with no help from me.
The method matters more than the outputs. The agent had to hunt for a real, evidenced pain (no claim allowed without a fetched source), generate competing pitches, run them through a tournament scored by an independent panel of judge agents, then subject the winner to adversarial attack — a sceptic agent, a red team, a claims audit — before building the brand, the website, the product and two rendered videos. Every decision logged, every question it would have asked me answered in the open.
Run one produced Unwritten, a studio that extracts what a solo expert actually knows — via Critical Decision Method interviews — and turns it into positioning that's provably theirs. Run two produced Bottle, a pressure-training studio for competent adults who choke under evaluation: interviews, exams, auditions. Both exist as complete artefacts — plans, brands, sites, films. Neither has been launched; that was the constraint, not the failure. The experiment was about how far autonomous building can go, and where it still needs a human.
Both premises, you'll notice, are my own ideas wearing company clothes. The agent found them in the knowledge system, because that's where I'd put them. Which brings me to —
The second brain that gets interviewed
My knowledge system is an Obsidian vault run by AI agents — but with one unusual rule: it doesn't just ingest documents, it elicits. A structured interviewer, built on the Critical Decision Method, periodically pulls out of my head the things I know but have never written down — the coaching eye, the decision rules, the cues. The reasoning behind that design is here: elicit, don't just ingest.
This site was written from that system. Every claim on it traces to a page in the vault.