How hedge‑fund manager Brian Kelly built Bracket22 with AI agents

How hedge‑fund manager Brian Kelly built Bracket22 with AI agents

THE MONEYWISE JOURNAL  ·  finance  ·  Thursday, September 10, 2026
How hedge‑fund manager Brian Kelly built Bracket22 with AI agents

By replacing a global staff that cost roughly $5 million a year with autonomous software agents, his annual operating expense fel…

By MoneyWise Guide

Brian Kelly, a former cryptocurrency hedge‑fund manager, launched Bracket22 as a fully AI‑driven trading firm in 2025. By replacing a global staff that cost roughly $5 million a year with autonomous software agents, his annual operating expense fell to between $30,000 and $40,000.

Key Takeaways

  • Kelly’s AI‑only model cuts labor‑related costs from about $5 million to under $40,000 per year.
  • Bracket22 runs three specialized agents—Steffi, Desmond, and Houston—to handle technical analysis, quantitative strategies, and overall coordination.
  • The firm trades only Kelly’s personal capital across cryptocurrencies, stocks, and commodities.
  • Kelly estimates a ten‑fold boost in productivity, and he sees AI as a multiplier for human teams rather than a wholesale replacement.
Metric Value Why it matters
Founder Brian Kelly Former crypto‑hedge‑fund manager who pioneered an AI‑only trading operation.
Company Bracket22 AI‑driven trading firm that operates without human staff.
Pre‑AI labor cost ~$5 million per year Shows the scale of traditional hedge‑fund payroll expenses.
Post‑AI labor cost $30,000–$40,000 per year Demonstrates the dramatic cost reduction achieved with AI agents.
Year AI testing began 2025 Marks the transition point from a human‑staffed fund to an AI‑only model.

From a Global Staff to a Virtual Team

Before embracing AI, Kelly maintained a team of seven to eight employees spread across several continents, with many based in New York. He told CNBC that salaries, health benefits, office space, and bonuses pushed his payroll “well into the millions of dollars per year.” Adding compute costs, the total labor‑related expense hovered around $5 million annually.

“Now, when I'm using AI, I run somewhere around $30,000 to $40,000 a year, total. And that's with every AI agent, that's with all my compute, that's with everything I need to completely replicate a hedge fund … with AI,” Kelly said.

By 2025, Kelly had shut down his cryptocurrency hedge fund and redirected the capital into Bracket22, a firm that trades his own money across three asset classes—cryptocurrencies, equities, and commodities—using only software agents.

The Architecture of Bracket22’s AI Agents

Kelly introduced three distinct bots, each designed to specialize in a single function:

  • Steffi – Performs technical analysis, scanning charts and identifying patterns that could signal entry or exit points.
  • Desmond – Executes quantitative strategies, applying statistical models to generate trade signals.
  • Houston – Acts as “mission control,” integrating the outputs of Steffi and Desmond and managing order execution, risk limits, and overall workflow.

Kelly emphasized that isolating each agent’s role allows the system to produce an “unbiased view” of market conditions. Human judgment still enters at the final decision stage, where Kelly reviews the AI‑generated recommendations before committing capital.

Productivity Gains and Cost Efficiency

Kelly estimates that the AI suite makes his operation at least ten times more productive than the previous human‑staffed model. The reduction in payroll also frees up capital that can be redirected toward additional compute power or new strategy development.

He envisions a hybrid future: “If you take a staff of 100, [with AI] you've got a staff of a thousand… 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.” This perspective aligns with broader industry discussions about AI as a productivity multiplier rather than a wholesale replacement.

Wall Street’s Parallel AI Experiments

Bracket22’s model reflects a growing trend among major banks. JPMorgan Chase’s CEO Jamie Dimon announced in February that the firm is redeploying staff to work alongside AI agents capable of operating autonomously for extended periods. Morgan Stanley is also shifting certain functions to AI. However, not all industry leaders are fully convinced; a Goldman Sachs partner recently warned that over‑reliance on AI could erode bankers’ analytical reasoning.

Potential Risks and Limitations

While the cost savings are striking, the AI‑only approach carries inherent risks. Automated systems can misinterpret market anomalies, and the lack of human oversight may delay the detection of model drift or data quality issues. Moreover, regulatory frameworks for AI‑driven trading are still evolving, leaving firms like Bracket22 navigating uncertain compliance terrain.

Why it matters

Kelly’s experiment demonstrates that AI can dramatically shrink operating expenses for boutique trading firms, potentially reshaping the economics of hedge‑fund management. For investors, the model raises questions about the sustainability of AI‑only strategies, the robustness of risk controls, and the future role of human expertise in finance. Watching how regulators and larger institutions respond will be key to understanding whether AI‑driven funds become mainstream or remain niche experiments.

Frequently asked questions

What is Bracket22?

Bracket22 is a trading firm founded by Brian Kelly in 2025 that uses only autonomous AI agents to trade Kelly’s personal capital across cryptocurrencies, stocks, and commodities.

How much does Bracket22 spend on labor after adopting AI?

The firm’s total annual cost for AI agents, compute, and related expenses is between $30,000 and $40,000.

Which AI agents does Kelly use and what do they do?

He employs three bots: “Steffi” for technical analysis, “Desmond” for quantitative strategies, and “Houston” as the central coordinator that integrates signals and manages execution.

How does Kelly’s cost reduction compare to traditional hedge‑fund expenses?

Prior to AI, Kelly’s labor‑related costs—including salaries, benefits, office space, and bonuses—were roughly $5 million per year, illustrating a reduction of over 99% after the switch.

Is the AI‑only model intended to replace human traders entirely?

Kelly views AI as an augmentation tool: while his firm runs without staff, he believes AI can make existing human teams up to ten times more productive, suggesting a hybrid future rather than full replacement.

Bottom Line

Brian Kelly’s Bracket22 proves that a hedge‑fund can operate with a payroll under $40,000 by delegating every trading function to specialized AI agents, cutting traditional labor costs by more than 99% while claiming a tenfold boost in productivity. The model highlights both the cost‑saving potential of AI in finance and the need for careful oversight as the industry balances automation with human judgment.

This is educational information, not personalized financial advice.

Reporting based on: original source.