A company that builds companies.

We build human-centered software at the frontier of AI — then back it, and the founders behind it, for the long run.

Principles we won't compromise.

If it doesn't solve a real problem, it doesn't ship.

Fast software hands people back their time and attention.

Simple systems are the ones people can rely on.

We build and invest to compound, not to chase the news.

We don't bolt AI on at the end. We design with it — and around the people it serves.

enum Tree { Leaf, Node(Box<Tree>, Box<Tree>), } fn size(t: &Tree) -> u32 { match t { Tree::Leaf => 1, Tree::Node(l, r) => size(l) + size(r), } }

Considered interfaces Fast by default Simple to rely on

Everything starts as an experiment. The few that earn it become companies — two are live today.

A shared source of truth for coding-agent teams. Human supervision, durable context, reviewable handoffs.

Deep research at scale. Define a schema, point it at thousands of entities, get validated structured data back.

Returns fund what we build next — the studio feeds the balance sheet, and the balance sheet funds the studio.

Human-centered software at the frontier of AI.

Ship it, and let it earn its place with real users.

What works becomes its own company.

The returns fund the next thing we build.

We invest our own capital — earned, not raised.

Long horizon

Concentrated positions in durable businesses with honest fundamentals and room to compound.

Early conviction

Backing founders building things we'd want to use ourselves, in domains we understand deeply.

Two ways in — whether you're building something, or have something to back.