London · Fontainebleau
AI OS
One obsession drives all of this: everything I read, say, meet and build becomes a permanent artefact - something I can reference for the rest of my life. It all lands in a single vault of plain text, and a small computer at home reads, sorts, drafts and files it while I sleep. Here is how the pieces fit together.
The vault itself
Everything lives in one folder of plain text. No proprietary app, no lock-in, nothing that dies when a company does - an artefact is only permanent if you still own it in fifty years.
For example: Any editor on any machine can open every note I have ever written.
Capture
The front door. Anything I notice, say, photograph or get sent lands in the vault with almost no effort - because the moment capture feels like filing, you stop capturing, and the record gets holes in it.
For example: I ramble into my phone; minutes later a tidy, structured note is sitting in my vault.
The autonomous engine
A small Mac Mini runs around the clock at home. It watches my inboxes, sorts my tasks, ranks what is urgent and nudges me - the difference between a pile of notes and a system that runs itself.
For example: Overnight, a rough capture becomes a researched, prioritised ticket sitting in the morning queue.
Reading & research
Articles, podcasts, transcripts, questions - the system reads them and writes the note I would want to find in five years, linked into everything else I know. Written once, referenced forever.
For example: I paste any article and get back a structured note - key points, quotes and links to what I already know.
Study & INSEAD
A whole pipeline for my MBA at INSEAD - downloading materials, turning lecture PDFs into structured notes, generating revision podcasts, tutoring me for exams. Two years of study, permanently searchable.
For example: A lecture handout goes in; a study note comes out with every chart and diagram preserved.
People
A private CRM built from plain notes. Every person I deal with has a card that updates itself each time we interact - so the record of a relationship is always better than my memory of it.
For example: Send one email and the card quietly updates - last contact, interaction count, what changed.
Writing & sending
The system knows how I write. It drafts emails, messages and posts in my voice, checks them against my own style rules, and can send them - every send becomes part of the record.
For example: I say 'write to X about Y'; it reads our full history, then drafts something I would actually send.
Diary & life admin
My calendar is wired straight in - the system reads it live, creates events, finds meeting slots and turns a fuzzy place name into a real address. Even the tennis court bookings run through it.
For example: Ask for lunch with a friend next week and it walks the diary, proposes times that work and creates the event.
Things I've built
The same setup that runs my life also builds real software. These are live products and ventures that grew out of it - proof the system compounds instead of just accumulating.
For example: An AI that drafts email replies for a restaurant overnight, learning how the owner writes from a playbook.
Making it look good
Not just text. The system produces decks, one-slide briefs, recipe cards and whole pages, and it remembers brand identities - so the tenth deck matches the first.
For example: Describe a deck and it renders a properly editable PowerPoint in whatever brand I ask for.
How I build
Building real software is the one place I slow down. A gated, reviewed process runs the work - AI agents build, verify each other, and nothing is committed until the checks pass.
For example: A foreman AI plans the work, worker AIs do the tasks, and a sceptical verifier attacks the result before anything is kept.
The foundations
None of this works without the plumbing: the same always-on machine, a memory that persists across every conversation, safety checks that run before anything destructive, and connections out to the wider world.
For example: A new conversation starts already knowing every project, every preference and every past mistake.
Beyond one person
Everything above is single-player: one life, one vault, one agent that can see all of it. The question I keep coming back to: how would this work for a team, or a whole company - every email, meeting and decision flowing into one shared memory that keeps itself current? Storage is easy; the hard part is a brain that reads each new thing against everything it already knows.
For example: Ask the shared brain a question and it answers with something a teammate learned last Tuesday.