A former hedge fund manager has built a trading firm that runs entirely on AI agents, slashing his annual labor costs from roughly $5 million to between $30,000 and $40,000. Brian Kelly, who previously ran a cryptocurrency hedge fund, created Bracket22 to test how far agentic AI can replace - or augment - a traditional finance team.
"I used to have about seven or eight employees all around the world. A lot of them were based in New York," Kelly told CNBC. "Between their salaries and compute and healthcare and everything like that, my payroll was well into the millions of dollars per year."
Bracket22 invests Kelly's own capital across cryptocurrencies, stocks, and commodities. The firm is a concentrated experiment in what happens when AI handles the analytical work that once required a global staff.
One manager, multiple AI specialists
Kelly introduced several AI agents, each with a defined role. "Steffi" handles technical analysis. "Desmond" runs quantitative strategies. "Houston" serves as mission control, synthesizing outputs from the other agents.
"I've crafted each of these agents to be a specialist in their field," Kelly said. "I wanted to isolate them and I wanted to get their unbiased view on what I'm doing." He then applies his own judgment to make the final trading decision.
Kelly estimates he is "at least 10 times more productive" with the agents. But he sees the larger opportunity in augmenting existing workforces rather than simply replacing them. "If you take a staff of 100, [with AI] you've got a staff of a thousand," he said. "It's not necessarily just, hey, you can replace everybody with AI agents. You can make your existing employees at least 10 times - maybe more - more productive."
Wall Street's broader AI shift
Bracket22 is an extreme case, but larger institutions are moving in the same direction. JPMorgan Chase CEO Jamie Dimon said in February that AI was already reshaping his workforce, with "huge redeployment" plans underway. The bank intends to launch AI agents later this year that can work autonomously for hours.
Morgan Stanley is also routing work to AI systems. Not everyone is moving at full speed, however. A Goldman Sachs partner recently warned about the risk of letting AI erode bankers' reasoning skills - a concern that echoes Kelly's emphasis on keeping a human in the loop for final decisions.
For management professionals watching these developments, the AI for Finance landscape is shifting from theoretical pilots to real cost structures. The numbers from Bracket22 offer a concrete benchmark for what's already possible.
Why this matters for managers
Kelly's experiment surfaces a question every manager will need to answer: not whether AI can replace roles, but how to redesign workflows so that human judgment sits at the right decision points. The 10x productivity claim is his estimate, not an industry average, but the cost differential - $5 million down to $40,000 - is a specific, measurable outcome. Managers evaluating AI for Management should treat that figure as one data point in a larger conversation about team structure, not a universal target.
The risk flagged by Goldman Sachs is equally concrete. If junior analysts stop building reasoning skills because AI handles the analysis, firms may save on headcount today and pay for it in judgment gaps tomorrow. The firms that get this right will be the ones that define which decisions AI informs and which ones humans still own.
Your membership also unlocks:
