Brian Kelly, a former cryptocurrency hedge fund manager, established Bracket22 as a trading firm operated entirely by agentic artificial intelligence. This shift drastically reduced his operational expenses; while previous staffing, including salaries, healthcare, and office space, cost approximately $5 million annually, the current AI infrastructure requires only $30,000 to $40,000 per year. The firm trades cryptocurrencies, stocks, and commodities using Kelly's own capital.
The setup features specialized AI agents such as "Steffi" for technical analysis, "Desmond" for quantitative strategies, and "Houston" for mission control. Kelly retains final decision-making authority but reports being at least ten times more productive than before. His approach mirrors broader Wall Street trends, where institutions like JPMorgan Chase and Morgan Stanley are integrating autonomous AI agents, though some firms remain cautious about potential impacts on human reasoning skills.
Kelly’s transition highlights a significant efficiency gain in financial operations, demonstrating how agentic AI can replace traditional labor structures with low-cost computational alternatives. By reducing fixed overheads to a fraction of their original size, small funds can potentially achieve scalability previously reserved for larger institutions with substantial tech budgets. This model suggests that the primary value of AI in finance may lie not just in automation, but in radically lowering the barrier to entry for sophisticated trading infrastructure.
However, the reliance on AI for core analytical functions introduces operational risks regarding model bias and system integrity. While Kelly emphasizes human oversight for final decisions, the industry must still address concerns raised by peers like Goldman Sachs regarding the erosion of fundamental reasoning skills among professionals. As major banks deploy autonomous agents, the market structure will likely see increased pressure on firms to adopt similar technologies to remain competitive, shifting the focus from headcount to algorithmic capability.
