1 hour ago
2
Prediction markets are getting complicated fast. Kalshi’s solution: build an AI agent to keep up.
The CFTC-regulated exchange has developed an internally built AI agent called Harrison, designed to help design and stress-test the prediction market contracts that underpin its rapidly growing platform. Co-founder Luana Lopes Lara described Harrison as a critical tool for managing the operational load of a platform now handling millions of daily wagers tied to everything from elections to sports to award ceremonies.
What Harrison actually does
Harrison isn’t a consumer-facing product. It’s an internal tool built to support Kalshi’s workflows in several key areas.
First, contract design. Prediction markets live and die by the precision of their contract language. A poorly worded contract on, say, whether a specific candidate wins a primary can lead to resolution disputes, user frustration, and regulatory headaches. Harrison is designed to help stress-test these contracts before they go live, catching ambiguities and edge cases that human reviewers might miss under time pressure.
Second, news aggregation. When you’re running contracts on real-world events, staying on top of rapidly evolving news is table stakes. Harrison automates the ingestion and processing of relevant information, giving Kalshi’s team a faster read on developments that could affect active markets.
Third, competitor analysis. The prediction market space has gotten significantly more crowded, and understanding what rivals are offering, and how they’re structuring their contracts, matters for staying competitive.
The scale problem Kalshi is trying to solve
Kalshi was founded in 2018 by Tarek Mansour and Luana Lopes Lara. It became the first CFTC-regulated exchange for event contracts, a distinction that gave it a significant early-mover advantage in the US market.
That advantage has translated into serious growth. The platform has reached valuations as high as $22 billion and now counts over 5 million users. Those numbers bring a scale problem that most prediction market platforms haven’t had to deal with before.
When you’re running thousands of contracts simultaneously across elections, sports events, and entertainment categories like award ceremonies, the operational burden compounds quickly. Each contract needs precise resolution criteria, needs to be monitored against real-world outcomes, and needs to be checked against regulatory requirements. Multiply that by the volume Kalshi is processing, millions of daily wagers according to Lopes Lara, and you start to understand why an AI assist makes sense.
AI in prediction markets: broader context
Kalshi previously explored a partnership with xAI, Elon Musk’s AI venture, though the specifics of how that collaboration evolved remain unclear. The development of Harrison suggests the company has decided that building in-house capabilities is at least part of the answer, rather than relying entirely on external AI providers.
No specific performance metrics for Harrison have been disclosed. We don’t know how many contracts it has reviewed, how many edge cases it has caught, or whether it has prevented any resolution disputes.
What this means for the market
For traders actively using Kalshi, the most immediate implication is potentially cleaner contracts. If Harrison works as intended, users should encounter fewer ambiguous resolution criteria and fewer disputes over contract outcomes.
The risk to watch: AI-assisted contract design is only as good as the data and rules it’s trained on. If Harrison’s stress-testing misses an edge case that leads to a high-profile resolution dispute, particularly on a politically sensitive contract, the reputational damage could be amplified precisely because the company touted its AI capabilities. In a regulated market, the CFTC will be watching whether AI-designed contracts meet the same standards as human-designed ones.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) · 