Hedge Fund Uses AI Agents Instead of a Traditional Team
Hedge fund manager Brian Kelly created Bracket22, powered by AI agents, and said its total annual cost is approximately $30,000 to $40,000, with final decisions made by a human.
What are the key facts?
- 1Bracket22 is operated entirely using AI agents
- 2Annual cost after adopting AI is approximately $30,000 to $40,000
- 3Kelly still makes final decisions using human judgment
- 4Kelly said AI agents increased his productivity at least tenfold
What happened?
Hedge fund manager Brian Kelly created a new trading company, Bracket22, operated using AI agents. Kelly previously had approximately seven to eight employees distributed around the world, with annual labor-related costs of about $5 million, including salaries, computing resources, health benefits, office space and bonuses. He now said that after adopting AI, the total annual cost, including all AI agents and computing resources, is approximately $30,000 to $40,000. Bracket22 invests only Kelly’s own money in cryptocurrencies, stocks and commodities. Kelly assigned specialized agents to different tasks: Steffi handles technical analysis, Desmond handles quantitative strategies, and Houston coordinates the work. He still relies on his own judgment and insights to make final decisions and said his productivity has increased at least tenfold after using the agents. The case reflects Wall Street institutions’ testing of AI in finance, while also highlighting concerns that AI could weaken professionals’ reasoning abilities.
What does this mean for cross-border sellers?
The Bracket22 case shows that AI agents can be assigned to tasks such as technical analysis, quantitative strategies and coordination to reduce repetitive-work costs and improve individual productivity, while final decisions remain with a human. Cross-border sellers can first assign customer service and pre-sales inquiries to AI, while retaining human review and avoiding handing all operational decisions to automation.
What should sellers do now?
- 1Select one type of frequent buyer question this week, organize standard answers, applicable conditions and cases requiring human escalation, and connect it to an AI customer-service test.AI Customer Service
- 2Create a separate list for pre-sales inquiries covering product specifications, compatible scenarios and prohibited responses so AI can handle initial routing of common questions.AI Presales Agent
- 3Sample AI-generated customer-service and pre-sales replies daily, record error types and update the knowledge base; retain human approval for replies involving pricing, refunds or abnormal orders.