aeqi vs Pi

Pi is one of the sharpest ideas in the agent space right now, and it's sharp by subtraction. Where most harnesses ship a 7–10K-token system prompt and a wall of tools, Pi ships a scaffold under a thousand tokens, four base tools, and skills that load only when needed. The bet is elegant: the model is already smart; get out of its way, and context economy beats a big framework. For a developer who wants a fast, cheap, transparent coding agent in their own terminal, on their own keys, Pi is an excellent answer.

aeqi is answering a different question. Not “how little scaffold does one agent need to write code,” but “what does it take for a Company to run itself.” aeqi is a Company OS: describe the business and a team of agents forms around it — roles, permissions, quests, a live website, and a memory that compounds. It isn't a coding loop with more features bolted on. It's the operating context a company needs so that agent work becomes durable execution instead of a one-off session.

The honest shape of the difference

These are not the same tool graded on one scale. Pi optimizes a single loop; aeqi runs an organization. A single loop is a feature when the job is “fix this bug” and a limit when the job is “keep this business coherent for a month” — because operations demand the things a minimal loop deliberately omits: memory that survives across sessions, several agents coordinating, roles and approval boundaries so authority is scoped. Pi's minimalism is a strength for coding and a constraint for operations. That's not a knock on Pi; it's the design working as intended.

And the trade cuts both ways. aeqi's richer scaffold does more, so it costs more and has more that can drift; Pi's economy is a real advantage on cost and reliability. Which structure actually wins at running a company — and at what price — is not something either of us should get to assert. It has to be measured.

So we're measuring it

That measurement is CompanyBench: a harness runs a store's back office over a run of shifts, every action graded by code against a golden — no model judging another model — on success, cost-to-run, and how long it keeps the books coherent before it drifts. Same model held across every entrant, so the score is the scaffolding. Pi and aeqi are both in the ring. Honestly: those runs haven't landed yet — the board says “queued,” the method is public and pre-registered, and we publish our own losses, aeqi included.

The verdict

Choose Pi if you want a minimal, fast, transparent coding agent at your terminal — it is one of the best expressions of that idea. Choose aeqi if you want a Company that runs itself: operations, a team, and memory, over time. Different tools for genuinely different jobs. The one place they truly overlap — can this scaffold run a business? — is the one place we refuse to guess. That's what the benchmark is for.