What if an organization was governed by prediction markets instead of management?
A futarchic organization is one where decisions are made by betting, not by authority. Instead of a manager deciding "use this tool" or "assign this task to that person," a market decides — and the market is accountable to measurable outcomes.
We're building this for AI agent teams first. A team of AI agents that write code, review pull requests, do research, and coordinate complex work. The agents operate in an economy: tasks have bounties, work has prices, predictions have stakes. The market governs who does what, how resources are allocated, and how the team evolves over time.
This isn't a new idea. Futarchy was proposed by Robin Hanson in 2000 as a form of government. Jeffrey Wernick actually ran a company governed by prediction markets in the 1980s. More recently, MetaDAO brought futarchy on-chain on Solana — using conditional markets on token price to govern DAO proposals. Combinator extended this with multi-option "quantum markets" for token parameter optimization. Optimism ran a futarchy experiment for grant allocation.
All of these govern human organizations making discrete decisions (pass a proposal, set a fee, allocate a grant). What none of them do is govern an autonomous agent team — an organization where the actors themselves are AI, the feedback loop is tight (hours, not weeks), and the market continuously steers operational behavior rather than voting on proposals one at a time.
Futarchy is hard to test with human organizations — too slow, too political, too many confounding variables. Agent teams are the perfect testbed:
- Fast iteration. An agent team can run hundreds of tasks per day. You get statistically meaningful data quickly.
- Measurable outcomes. Did the code pass review? Did the prediction come true? Did the change improve performance? Binary, verifiable, no ambiguity.
- Controlled experiments. You can fork a team configuration, run both versions, and compare. Try that with a human org.
- Real stakes without real risk. Start with internal currency. Graduate to real money when the mechanics are proven.
We're exploring several connected ideas. Each can stand alone, but they're strongest together:
| Idea | What it is | Status |
|---|---|---|
| Futarchic governance | Market mechanisms governing an agent team. The evaluation layer sits outside the system being evaluated. | Brainstorming |
| Agent economy | Agents earn and spend currency through specialized work. Supply and demand pricing. | Brainstorming |
| PR prediction markets | Prediction markets on whether a PR will be merged. Useful signal for maintainers. Standalone product. | Brainstorming |
| Scaffold evolution | Evolving agent prompts and configs via natural selection, driven by market fitness signals. No model retraining. | Research needed |
| Counterfactual evaluation | Measuring an agent's value by comparing outcomes with and without their contribution. | Research needed |
See the brainstorm index for detailed thinking on each.
Early stage. We're figuring out what to build first and writing things down as we go. The brainstorm folder is where the real thinking lives.
MIT