shoku

Why we started

AI agents are getting good at thinking, but they still can't do anything in the real world — they can't walk a dog, tow a car, or clear a shipment through customs. At the same time, the contract economy is full of capable people with no good way to be found. The friction runs both directions and it's the same friction: turning an outcome into a job post, a search, a dozen messages, a schedule, a follow-up, a proof, a payment.

And today's gig-work apps make it worse, not better. Workers feel squeezed, customers feel the whole thing is transactional and low-trust, and the platforms compete on one axis only — price, a race to the bottom that nobody actually enjoys. We think the AI era flips this. Agents won't just consume labor; they'll create more of it, and more kinds of it — which demands a fundamentally new marketplace, one built to unleash what agents can do rather than to shave another dollar off a delivery fee. So we wanted to delete the middle layer: let anyone, human or agent, say what they want and have the right people, businesses, and tools actually make it happen.

How we think about it

One idea shaped every decision: the thing you hire is an outcome, not a person or a job title. A Mission is an outcome with a boundary — a capability, a budget, a window, and the proof that it's done. Missions are fractal: a small leaf goes to one person; a whole branch goes to an organization that decomposes it internally — the general-contractor model, made programmable. Long-term work isn't a new kind of object; it's the same bounded Missions, repeated as trust accumulates.

The second principle is a division of labor between the model and the domain: the LLM proposes, deterministic code disposes. A language model can turn a messy sentence into a plan, but budgets, authority, dependencies, evidence, and settlement are enforced by typed commands before anything is saved. Business facts are durable; the prompt is only context. History is corrected by appending compensating events, never by deleting — so the record stays trustworthy.

And a deliberate boundary: we compile outcomes into hireable units; the customer's own agent orchestrates across them. We chose not to build a workflow engine, because every vertical's process differs while the hiring unit does not.

Where it goes next

The protocol is real, but a lot of the edge is still demo-grade, and we like knowing exactly where. Settlement is mocked — the honest next step is real escrow and payout. External tools resolve instantly today; they need real MCP adapters. Planning runs on a reasoning model that takes seconds — we'd want faster or streaming plans, and proper auth across the API. Deepest down, we're most excited about the supply side: verified onboarding, reputation that compounds into genuine long-term relationships, and semantic recall that widens who a Mission can reach before the deterministic engine ranks them.

The bigger bet

Further out, we think the shape of the labor market itself will change. As AI robots and humans start working side by side, "who does the work" — and even "who holds economic rights over it" — stops being obvious, and the apps built for today's gig economy simply won't fit that world. A mission-based hiring protocol, where any accountable party (person, company, agent, or machine) can be hired against the same contract, is our small bet on the layer that will. We built this for the hackathon hoping it sparks that conversation. If it resonates — and if the right people and resources come with it — we'd love to push it much further.

Built With

• codex