Man Group’s quant arm, Man Numeric, has begun using a new agentic AI system capable of autonomously generating, coding, and backtesting trading strategies — a significant step in automating the quantitative research pipeline at the world’s largest listed hedge fund, according to a report by Bloomberg.

The internally developed tool, dubbed AlphaGPT, is designed to mimic the research process of human quant analysts. It mines historical data, formulates rule-based trading signals, writes the corresponding code, and evaluates performance through backtesting. While human oversight remains essential, AlphaGPT has already produced several dozen investment signals that have been approved for live trading, according to senior portfolio manager Ziang Fang.

The deployment of this system marks one of the first institutional-scale uses of agentic AI — a class of large language model-based tools capable of performing multi-step tasks with minimal human intervention. Fang notes the move addresses the “information avalanche” facing today’s quants, with AI helping to navigate vast data sets and academic literature faster and more effectively than manual research alone.

While issues like hallucinations and inconsistent outputs persist, Man Numeric is iterating with caution.

“I wouldn’t call it autopilot,” Fang said, “because we have a lot of checkpoints in place just to make sure we have the proper guardrails on our investment process.” Still, the ambition is to build a more autonomous system that can self-verify its work and continuously refine ideas at scale.