The Wall Street Bot – How Hedge Funds Are Replacing Junior Analysts with Fine-Tuned LLMs

A managing director at a midtown hedge fund first showed me his “team” of refined models, but before he could explain anything, he laughed. He gestured to four open chat windows on a monitor, each named after a junior analyst who had departed the company within the previous eighteen months. “We didn’t fire them,” he remarked while stirring his coffee. “We just stopped backfilling.” Most of the names were jokes. However, the underlying workflow wasn’t.

The lower echelons of Wall Street have operated in this manner for decades. Employ intelligent twenty-two-year-olds from Princeton or Wharton, give them a Bloomberg Terminal, and let them work through pitch books until two in the morning. Many in the industry are reluctant to publicly acknowledge that the world is changing, albeit subtly and unevenly. As early as April 2024, the New York Times reported that AI tools could replace a large portion of entry-level white-collar work in finance. By mid-2025, Fortune was issuing similar warnings. Those stories don’t yet have dramatic numbers. It’s the mood.

The real events taking place on trading floors are more fascinating than what the headlines portray. LLMs are flagging fraud signals in thousands of filings, writing investment memos that are cleaner than what a second-year associate would produce on a Tuesday at one in the morning, and analyzing earnings calls within seconds of their release. The technology has been incorporated into almost every workflow, with the exception of one at Balyasny, JPMorgan, and smaller pod shops that are unknown to the outside world. trading on its own. It’s telling that real money still passes through human hands.

The reasons were made clear by Nof1’s Alpha Arena competition. Two weeks’ worth of US tech stocks and eight frontier models, each costing ten thousand dollars. About one-third of the capital in the portfolio was lost. Only six of the thirty-two result sets ended in the green. Grok executed 158 trades in response to a single prompt, while Alibaba’s Qwen executed 1,418. Variance of that type is not a characteristic. It serves as a caution.

However, the results of the SWOT analyses indicate otherwise. When six top LLMs were compared to analyst consensus on Deutsche Telekom, Daiichi Sankyo, and Kirby Corporation, the machines frequently revealed risks that the humans had overlooked. Performance increased by up to 40% with sophisticated prompting. Walking through these numbers gives us the impression that two distinct jobs are being evaluated simultaneously. investigation and implementation. The bots perform embarrassingly poorly at one and surprisingly well at the other.

It’s more difficult to predict what this will mean for the child who graduates in May. Although it was always difficult, the route from analyst to associate to vice president was clear. The first two rungs are now swaying. This quarter, according to a friend who oversees recruiting at a long-short fund, there are more requests for “AI-fluent” hires and fewer junior ones. Early in 2025, the Australian fund Minotaur Capital made headlines when it completely replaced its analyst bench with AI and reportedly outperformed. Nobody knows if that holds true for the entire cycle.

The cultural shift that lies beneath the technical one is difficult to ignore. The myth of the brilliant young analyst has always been promoted by Wall Street. If you take that away, the industry will need to give a different explanation. Perhaps the companies figure it out. Perhaps they don’t. The only honest thing to say as you watch this happen is that while the people on top of the pyramid are still asleep, the base of the pyramid is being rebuilt.