Kalshi Inc. has implemented an internal AI assistant, internally named Harrison, to help manage several operational tasks tied to its prediction market platform. The co-founder, Luana Lopes Lara, confirmed the development and described the agent's role in reviewing contract language and handling high volumes of daily wagers on event outcomes such as elections, sports contests and award ceremonies.
Kalshi's contracts rely on precise wording and clearly defined evidence sources to determine how wagers are resolved. To address that sensitivity, Harrison performs a set of recurring duties that include aggregating leading news items, surveying competitor product offerings and producing recommendations for the exchange's pipeline of potential markets. The agent also suggests where the company might concentrate rewards to encourage users to add liquidity.
According to Lopes Lara, the markets team is the primary non-engineering user group for Harrison within Kalshi's approximately 150-person organization. The underlying model for the agent is Anthropic's Claude, on top of which Kalshi has tailored the system to its internal requirements.
Operationally, Kalshi maintains more than 500 templates for potential markets that have passed review by the internal team. Each template is examined to determine how broadly it can be generalized to cover additional events, how it can be stress-tested and whether it satisfies user requirements.
The company has formalized an outcome-determination workflow in most cases. The process involves three steps: one markets-team member enters an outcome into the system, a second team member independently records their decision, and Kalshi's AI compares the two human inputs to see if they match while also checking against its own suggested response.
Those elements combine to position Harrison as both a quality-control and decision-support tool in Kalshi's operations, helping the exchange manage the nuances of contract language and the volume of adjudication tasks associated with millions of daily wagers.
Contextual notes
The facts presented here are limited to Kalshi's internal deployment and stated operational processes. Details about the agent's technical specifications beyond the use of Anthropic's Claude model, or additional metrics on performance impact, were not provided in the information available.