TL;DR:
- Kalshi developed an AI agent called Harrison to optimize contract drafting and internal processes within its prediction market.
- The agent, built on Anthropic’s Claude model, analyzes competitors, suggests new markets and verifies the resolution of active bets.
- In May, Kalshi recorded a monthly volume of nearly $18 billion, and in the first week of the World Cup it broke its weekly record with $5.1 billion.
Kalshi developed an internal artificial intelligence agent to optimize one of the most persistent problems in prediction markets: the precise drafting of the contracts on which bets are placed. The tool, known within the company as Harrison, was publicly confirmed by cofounder Luana Lopes Lara in a recent interview.
The agent is built on Anthropic’s Claude model and is used primarily by the company’s markets team, which has a total of 150 employees. Its daily functions include aggregating relevant news, analyzing competitors’ offerings and generating recommendations on which markets to list next or where to concentrate liquidity incentives for users.
Harrison Stress-Tests Every Kalshi Contract
Precision in contract wording has direct consequences on bets worth several million dollars. Kalshi experienced this firsthand when it resolved a market on whether a Netflix executive would mention “Warner Bros.” on an earnings call as negative, given that the executive said “Warner Brothers” instead of the abbreviated form. Cases like that illustrate why the company invested resources in systematizing the contract validation process.
Before Harrison, Kalshi hired debate champions from Yale University to stress-test the structure of its contracts. Today the company has more than 500 already-evaluated templates, and the agent automatically suggests which one to apply in each case, anticipates potential interpretation conflicts and indicates whether any amendment or additional certification is required.
Market resolution will also incorporate the tool: when it detects intense press coverage of a topic, Harrison sends alerts to the markets team with a list of contracts that may need determination. In the most complex cases, such as United States Supreme Court rulings, the process escalates to involve the company’s regulatory director.
Record Trading Volume
Growing demand places a significant operational burden on the platform. In May, Kalshi processed nearly $18 billion in notional volume, according to data compiled by users on Dune Analytics. The first week of this month’s soccer World Cup deepened that trend with a weekly record of $5.1 billion, driven by massive demand for sports betting.
