Bracket22’s AI-only hedge fund model cuts costs
Hedge fund manager Brian Kelly has built a trading firm around artificial intelligence agents rather than a traditional investment team, cutting annual labour-related costs from an estimated $5m to as little as $40,000, according to a report by CNBC.
Hedge fund manager Brian Kelly has revolutionized his trading firm by replacing a traditional investment team with artificial intelligence agents, significantly cutting annual labor costs, according to a CNBC report. Previously, annual labor-related expenses for the firm were estimated at $5 million, but with the implementation of AI agents, costs have been reduced to as little as $40,000 annually.
Kelly, a former cryptocurrency hedge fund manager, launched Bracket22 after experimenting with AI in 2025. The firm specializes in trading cryptocurrencies, equities, and commodities using capital provided by Kelly only.
At the core of Bracket22's operations are a network of specialized AI agents that handle tasks typically performed by analysts, traders, and quantitative researchers. These AI agents perform functions such as technical analysis, quantitative strategy development, and central coordination of various functions. For instance, "Steffi" is responsible for technical analysis, "Desmond" for quantitative strategies, and "Houston" as a central control system that oversees and coordinates the various activities of the AI agents.
Kelly has meticulously assigned specific responsibilities to each AI agent. The agents do not make decisions independently; instead, their outputs are reviewed by Kelly, who acts as the final decision-making authority, combining their insights with his own judgment. Despite reducing his workforce to just himself, Kelly claims that the AI agents have increased his productivity by at least tenfold.
He further argues that the true potential of this technology lies not in eliminating employees entirely but in augmenting the capabilities of existing investment teams.
Written by urgent.news from Hedgeweek's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.