The Hedge Fund Run Only by AI: How TradingAgents Works and What You Can Copy

I have been trading for more than ten years and I have seen plenty of "systems that beat the market". TradingAgents is different, not because it promises more, but because it shows how a team reasons. In this article we look at how it works, what the numbers really say, and what you can copy without writing a single line of code.

What TradingAgents is

The idea is simple. Instead of asking a single artificial intelligence whether to buy or sell, the authors rebuild a trading firm made entirely of AI agents, each with its own role. In the paper they write that they were inspired by the organizational structure of real trading firms.

The code is public on GitHub under the Apache 2.0 license, so anyone can study it and use it. Today the project is run by Tauric Research.

The team, role by role

In the current version of the code there are twelve agents:

1. Four analysts, working in parallel: fundamentals, social media sentiment, news and technical analysis.
2. Two researchers: the bull defends the buy, the bear tears it apart. A research manager acts as the judge.
3. The trader, who reads everything and proposes the trade.
4. Three risk analysts, one aggressive, one neutral and one conservative, who challenge everything again.
5. The portfolio manager, who approves or rejects.

Every step is written down with its reasons. And there is a journal: every decision is logged and, when the real result comes in, another AI writes down the lesson learned, which the portfolio manager rereads before the next decision.

Notice who speaks last. Not the smartest analyst, not the trader. The last word goes to whoever thinks about what could go wrong.

The paper's numbers, read properly

The authors tested the team on Apple, Google and Amazon, from January to March 2024:

• Apple: +26.6% versus -5.2% for buy and hold.
• Google: +24.4% versus +7.8%.
• Amazon: +23.2% versus +17.1%.
• Maximum drawdown between 0.9% and 2.1%, and a Sharpe ratio of 8.21 on Apple.

It looks like a machine built to beat the market. But the test lasted three months, and the authors explain it was that short because of the cost of making all those agents reason. On the Sharpe ratio they write that it goes beyond the range they expected (for them, above 2 is very good and above 3 is excellent) and attribute it to the few pullbacks in that period.

The project's README is clear: TradingAgents is built for research and is not financial, investment or trading advice. And a study published in May 2026, "The Alpha Illusion", names TradingAgents among the systems whose reported returns should not be taken as proof that they work in the real market.

Who profits, if it's free

Three of the four authors also work for Tauric Research, the company that runs the project today and whose website offers trading solutions and access for businesses. There is nothing wrong with that: it is how plenty of tech companies work, you give away the code and you sell the service. It is the same reminder as the first AI millionaire: when something is free, ask who profits.

Three rules you can copy starting tomorrow

1. Always look for the opposite opinion. If you're the bull, ask the AI to play the bear.
2. Separate whoever analyzes from whoever decides. Write the analysis first, then decide with a cool head.
3. Risk speaks last, and it can say no. If the risk doesn't add up, the trade doesn't happen.

The team prompt (free)

I turned the team's method into a prompt ready to copy into ChatGPT, Claude, Gemini or Grok. It doesn't ask for signals: it makes you reason in 4 separate phases.

1. Analysis: four separate analysts (fundamentals, news, sentiment and chart), facts only, with the date of the data and what is not known.
2. Bull vs bear: the 3 best arguments for and the 3 against, and a judge who says which are stronger.
3. Risk check: three voices (aggressive, neutral, conservative), with the conservative voice holding veto power.
4. Written decision: 5 lines max, with the risk and the condition that invalidates the idea.

You'll find the full text for free on the prompts page: leave your name and email and copy it right away.

Sources

• TradingAgents paper, arXiv 2412.20138
• Official TradingAgents repository on GitHub
• Tauric Research
• The Alpha Illusion, arXiv 2605.16895
• Trendshift, repository page

The full story is told in video on my YouTube channel.

This article is for educational and informational purposes only and is not financial advice. Trading involves a high risk of losing capital.