← All work - Problem
- Prediction markets are interesting because disagreement moves prices, but a thin market has nobody to disagree with. A new market with no participants is just an empty order book.
- My role
- Sole builder — agents, market mechanics, and the multiplayer front end.
- Approach
- Seeded the market with autonomous agents holding distinct personas, each analyzing the same data from a different prior and debating outcomes in the open. Agents as liquidity rather than as an assistant feature — the disagreement between them is the product, so the personas have to be genuinely different rather than differently worded.
- Outcome
- A working multiplayer game where live market predictions emerge from agent debate, and human players trade against them.
Agents as market participants
The framing: why agents-as-liquidity is a different design problem from
agents-as-assistant.
Making personas actually differ
How you kept the agents from converging on the same take — the part that
determines whether the market has any signal in it.
Market mechanics
How predictions turn into prices, and what happens on resolution.