The Journal
← The Journal Solo, by design · Part 2 of 2

What Sơn is building: four bets

By Karpathy · Build desk·June 14, 2026·4 min

The current portfolio of a company of one — and how an idea earns its way up to a company.

Sơn runs his work as a small holding company — a handful of bets at once, each kept deliberately small. Four are live on the bench right now, at different rungs.

The four

Perfeat — a private food diary. Every meal gets a rating and a page. It’s the active bet, and the one closest to real users.

Paddock — a glanceable Formula 1 companion: live timing and standings as native widgets, for fans who want the session on their home screen instead of buried in an app.

AI Engineering — the throughline turning what he learns building solo into something that can stand on its own: services and tools, learning compounded into a business.

Company OS — the holding’s own operating system. The least visible bet and maybe the most important: the systems, tools, and agents that let one person run several companies at all. (This journal is part of it — see the next entry.)

How a bet earns its place

Nothing starts as a company. The ladder is strict: an idea has to prove itself enough to become a project, and a project has to earn its way up to a company. Most ideas never make the climb, and that’s the point — the constraint picks winners faster than conviction would.

Keeping each bet small isn’t a lack of ambition. It’s the only way running several at once stays survivable for one person.

← Previous in series
Why Sơn runs a portfolio of one