Bracket22 Cuts Labor Costs Using AI Agents

Brian Kelly built hedge fund Bracket22 around a network of specialized artificial intelligence agents and reduced annual labor-related costs from about $5 million to roughly $30,000-$40,000, while retaining human judgment as the final decision-maker.

Kelly launched Bracket22 after experimenting with AI in 2025. The firm trades cryptocurrencies, equities and commodities and is funded solely with his personal capital.

He organized the business so software handles tasks that would normally require analysts, traders and quantitative researchers. A set of purpose-built agents performs discrete functions and feeds outputs to Kelly, who makes the final trading calls.

Before the change, Kelly employed seven or eight people across multiple locations, including New York. After adding salaries, health care, technology, office costs and bonuses, he estimates annual labor-related expenses reached about $5 million. He now estimates the AI infrastructure, including cloud computing and the agents, costs $30,000-$40,000 a year.

He named the agents “Steffi,” “Desmond” and “Houston,” describing their roles: “Steffi focuses on technical analysis,” “Desmond works on quantitative strategies,” and “Houston acts as a central control system coordinating the different functions.” He described the arrangement and noted that “Human judgement remains the final decision-making layer.”

Kelly estimates working with the agents has made him at least ten times more productive. He expects the technology may expand the capacity of existing investment teams rather than eliminate staff entirely.

The model changes the cost structure for a small, privately funded trading operation and raises questions about how much of a traditional investment workflow can be automated. Kelly previously led a cryptocurrency hedge fund, and Bracket22 continues to trade across multiple asset classes.