Brett Harrison, the CEO of trading technology firm Architect and a veteran of Jane Street and former president of FTX US, argues that large language models are not up to the task of building effective high-frequency trading systems. In his view, human expertise remains irreplaceable when it comes to designing the algorithms that power modern markets.
Why Harrison is skeptical of LLMs in HFT
Harrison, who spent years at quantitative trading firm Jane Street before leading FTX US, says that the current generation of large language models simply cannot handle the demands of high-frequency trading. HFT systems need to make split-second decisions based on rapidly changing data. LLMs, for all their power in processing text and generating responses, lack the precision and speed required for that environment.
He didn't mince words. The technology, he argues, is not a substitute for the deep domain knowledge and careful engineering that goes into building a competitive trading system. While LLMs have shown promise in areas like sentiment analysis or summarizing news, they fall short when it comes to the core logic of trading strategies.
The enduring role of human expertise
Harrison's point is not that AI has no place in finance. It's that the hype around LLMs has outpaced their actual capabilities in certain high-stakes applications. Building a high-frequency trading system requires understanding market microstructure, latency constraints, and risk management in ways that a language model trained on internet text cannot replicate.
Human engineers, he says, bring judgment and experience that machines still lack. They can spot edge cases, adapt to new market conditions, and make nuanced decisions about when to override an algorithm. That human touch, in Harrison's view, is what separates a working system from a failing one.
His comments come as the financial industry pours billions into AI research. Banks, hedge funds, and trading firms are all experimenting with large language models for everything from customer service to trade execution. But Harrison's skepticism serves as a reminder that not every problem is best solved by the latest AI trend.
For now, Architect continues to build its trading technology the old-fashioned way: with human engineers writing the code and making the key design decisions. Harrison's message is clear: don't expect LLMs to take over the trading floor anytime soon.
