This Hedge Fund Manager Replaced His Entire Staff With AI Agents and Cut Costs From $5M to $40K

Brian Kelly ran a hedge fund with seven or eight employees and a payroll in the millions. His new firm, Bracket22, runs on AI agents for $30,000 to $40,000 a year, all in. This CNBC Fast Money segment tours his Miami operation, where named agents handle technical analysis, quantitative strategy, and coordination while Kelly makes the final calls. The on-screen graphic puts the comparison bluntly: $5M a year for human staff before, $30-40K for AI bots and compute now.

The Cost Math

Kelly's prior fund employed seven or eight people around the world, many in New York. Between salaries, compute, and healthcare, payroll ran well into the millions per year. CNBC's graphic, sourced to Bracket22, pegs the old annual cost at $5M against $30-40K for the new setup. That figure covers every AI agent and all the compute Kelly says he needs to replicate a hedge fund. The reduction is roughly 99 percent, and it converts a fixed headcount cost into a small operating expense.

A Staff of Named Agents

Kelly rebuilt his old org chart as software. Each bot has its own memory and a defined job:

Steffi Graff runs technical analysis. Her standing instruction is to be the best technical analyst in the world, and Kelly shows chart work she produced that day.
Desmond owns quantitative strategies and returns a quant point of view on any question.
Houston acts as mission control. The specialist agents report up to Houston, and Houston relays the synthesis to Kelly. On camera, Kelly asks Houston to pull up Desmond's charts by voice.

The footage shows Kelly working a wall-mounted display in a home office, scrolling an agent chat interface with embedded charts and drawdown plots. He calls the agent-to-agent channel "the corporate brain," and the agents monitor markets for him 24 hours a day.

The Human Stays in the Loop

The agents do not trade on their own. Each one brings its analysis to Houston, Houston relays it to Kelly, and Kelly applies his own judgment to make the final decision. The structure mirrors how analysts feed a portfolio manager at a traditional fund. The labor got automated. The decision rights did not move.

Augmentation Over Replacement

Kelly estimates he is at least 10 times more productive with his agents, and he frames the broader lesson as multiplication rather than substitution: a staff of 100 becomes a staff of 1,000. He says the point is not that you can replace everybody with AI agents, but that existing employees can get 10x or more productive. That framing matters coming from someone who did fully replace his staff, because it positions Bracket22 as the extreme end of a spectrum most firms will enter through augmentation.

Why CNBC Is Covering This Now

The anchor sets up the piece with context: brokerages from Robinhood to Webull are letting retail clients make AI-powered investments, and professional money managers are leaning on agentic tools too. Kelly, a longtime Fast Money contributor who previously ran a cryptocurrency hedge fund, is an early data point for what a fully agentic investment firm looks like in practice. Bracket22 trades crypto, stocks, and commodities with Kelly's own capital, which sidesteps the client-money questions a fund with outside investors would face.

Key Takeaways

Bracket22 replaced a seven-to-eight person hedge fund staff with AI agents, cutting annual costs from roughly $5M to $30-40K.
Each agent has a named role and its own memory: Steffi Graff on technical analysis, Desmond on quant strategy, Houston as mission control.
Agents monitor markets 24/7 and feed synthesized views to Kelly, who keeps final decision authority.
Kelly claims at least 10x personal productivity and argues the real opportunity is making existing employees 10x more productive, not firing them.
The firm trades only Kelly's own capital, which makes it a clean test case rather than a template for funds managing client money.