Bracket22: the hedge fund Brian Kelly runs entirely on AI agents

Brian Kelly, who ran a cryptocurrency hedge fund and was a trader on CNBC's 'Fast Money', shut down his crypto fund in early 2025. Months later he began experimenting with artificial intelligence and ended up founding Bracket22, a trading firm that, by his own account, runs entirely on agentic AI. Bracket22 invests only Kelly's own capital and trades cryptocurrencies, stocks and commodities.

The economic contrast Kelly draws is the most striking figure in the piece. At his previous fund he had seven or eight employees spread around the world, many of them based in New York, and he estimates that between salaries, computing, healthcare, offices and bonuses he spent around $5 million a year on labor costs. With Bracket22, he says total spending — covering all the AI agents, all the computing capacity and any other resource needed to replicate a full hedge fund — runs between $30,000 and $40,000 a year.

The system relies on several specialized agents with names of their own: 'Steffi' handles technical analysis, 'Desmond' quantitative strategies, and 'Houston' acts as a command center that integrates the others' work. Kelly explains that he designed each agent as a specialist isolated within its own discipline precisely to get an 'unbiased' view from each one of what he is doing, and that he then applies his own human judgment to make the final call.

As for results, Kelly says he feels 'at least ten times more productive' than before. It is worth stressing that this is a personal estimate from the founder himself, with no fund performance figures or verified third-party data in the article to back it up.

The case fits a broader trend CNBC has documented on Wall Street: JPMorgan Chase CEO Jamie Dimon said in February that the technology is already reshaping the bank's workforce, with plans for a 'massive redeployment' of employees and the launch this year of AI agents able to work autonomously for hours. Morgan Stanley is also shifting part of its work to AI systems. Not every adoption comes without reservations: a Goldman Sachs partner recently warned of the risk that AI erodes bankers' reasoning skills, a note of caution that contrasts with Kelly's enthusiasm.

Despite having replaced his own human team, Kelly qualifies that he does not see AI primarily as a replacement tool but as a way to expand capacity: 'if you have a staff of 100 people, with AI you have a staff of a thousand,' he said, insisting that the real opportunity lies in multiplying the productivity of existing employees and not merely in replacing them. It is a perspective that softens the story's central message — a hedge fund with no human employees — but it does not dispel the underlying questions the case raises: what regulatory or risk oversight exists over a fund run this way, what happens when the agents get things wrong in a coordinated fashion, and how replicable this model is outside a fund that manages only its own capital.