Building a 6-Agent AI System to Debate Trade Decisions

The author built a 24/7 AI trading system that uses 6 specialized agents to analyze and debate each trade decision. The system identified a mistake where it chose to sit out a trade that ended up rallying 6% that day.

Details

The author built a 6-agent AI trading system in just 16 days, with the help of the Claude AI assistant. The system runs 24/7 and monitors over 87 symbols across the US, China, and Hong Kong markets. Each trade decision goes through a debate process where 6 specialized agents analyze the symbol from different perspectives - technical, fundamental, sentiment, fund flows, macroeconomics, and contrarian views. The system doesn't just average the agents' scores, but explicitly identifies contradictions between them, as these contradictions are seen as valuable signals. For example, if the technical analysis is bullish but the fundamentals are bearish, it flags a potential value trap. The system then makes the final trading decision, which can even reject trades where the agents agree. On March 31, the system analyzed Amphenol (APH) and the 6 agents were split 3-3 on whether to buy or sell, so the system chose to sit it out. However, APH ended up rallying 6% that day, which was the system's first real mistake. The author discusses the lessons learned from this experience and the importance of detecting contradictions in the analysis.