aeqi is built by an aeqi Company

Every team building agent infrastructure makes the same claim: our agents do real work. Almost none of them can show you the ledger. So here is ours. The company that builds aeqi is a Company running on aeqi — same runtime, same primitives, same lifecycle we sell. This post is the receipts.

The claim, precisely

Dogfooding usually means something weak: the team has the app installed, someone checks the dashboard on Fridays. This is not that. There is no second system behind the curtain — no private issue tracker where the real work happens, no wiki where the real decisions live. The Company that builds aeqi runs on the four primitives every customer gets: agents occupying roles, quests carrying the work, ideas holding the memory, events moving state.

When something needs to be built, it is filed as a quest before work starts. When an agent picks it up, the quest moves; when the work merges, the quest closes with a result a stranger could audit. When a decision gets made — positioning, architecture, a trap discovered at 2am — it is stored in the Company's memory, where every agent that touches that territory later will retrieve it. The company is not using a tool. The company is the deployment.

The receipts

As I write this, the ledger stands at 2,262 quests filed and 1,624 closed. In the runtime repository — roughly 300,000 lines of Rust — 2,355 of the 5,189 commits carry an agent co-author line. For the site you are reading right now, it is 382 of 720. These are not demo numbers generated for a screenshot. They are the operating history of a company that has been building a real product this way for months.

The texture matters more than the totals. This morning, three agents ran an adversarial audit of the runtime's agent loop in parallel — each attacking a different subsystem — and their findings were consolidated, verified against the code line by line, and turned into the next sprint before lunch. Last month, when our positioning changed, the correction was stored once as an idea in the Company's memory; every agent that has written a line of public copy since has retrieved it and complied. Nobody circulated a memo. The memory is the memo.

And this post itself: filed as a quest, drafted by an agent against the Company's content canon, previewed on a branch, accepted by a human, merged through the same lifecycle as every other unit of work. The quest closed when you could read it.

What it changes about the product

A company OS built any other way is a hypothesis. The team imagines what an agent-operated company needs, ships it, and waits for someone else to hit the wall. We hit the wall first, every day, in production, under deadline — and the walls are almost never where the roadmap thought they were.

Quests grew their lifecycle discipline — file before work, close with an auditable result — because untracked agent work turned out to be unreviewable agent work. Memory grew its retrieval habits because an agent that recalls a stale decision and acts on it is worse than an agent that knows nothing. Roles grew real authority boundaries because the first time an agent needs production access is not the moment to invent a permission model. None of this came from a product spec. It came from operating pain, felt by the people with the power to fix it, in the same week they felt it.

That loop is the moat. Not the features themselves — features can be copied — but the thousands of closed quests of operating history that decided which features matter and which are theater.

What it proves, and what it doesn't

Honesty about the claim is part of the claim. The receipts do not show a company running itself while the founder sleeps. They show something more useful: a real division of labor that holds up under load. A human sets direction, decides what ships, and owns every production touch. Agents execute — they implement, audit, verify, write, and remember. The interesting number is not "how much did the agents do." It is how far one person's direction now reaches: one founder, thousands of closed quests, a runtime of roughly 300,000 lines of Rust, and a product that ships every week.

That is the actual promise of a programmable company: not zero humans, but a Company where human judgment is the scarce input and everything downstream of it compounds — into code, into memory, into a ledger you can audit.

The homepage says aeqi runs aeqi. This is what that sentence means. Not a slogan — a ledger. And when you launch a Company of your own, you get the same runtime we bet ours on.