Experts say artificial intelligence-driven trading has yet to reach the “iPhone era,” where everyone carries algorithmic reinforcement learning portfolio managers in their pockets, but something similar is coming.
In fact, the power of artificial intelligence comes into play when faced with dynamic, adversarial trading markets. Unlike an AI agent that learns through the endless circuits of a self-driving car to accurately recognize traffic signals, no data and model can predict the future.
This makes perfecting an AI trading model a complex and arduous process. Success is often measured by profit and loss (P&L). But advances in custom algorithms are giving rise to intelligent agents that continuously learn to balance risk and reward in the face of multiple market conditions.
Michael Sena, chief marketing officer at Recall Labs, said allowing risk-adjusted metrics like the Sharpe ratio to inform the learning process can exponentially increase the complexity of the test. Recall Labs is a company that operates around 20 AI trading arenas, where the community submits AI trading agents who compete over four to five days.
“As they scan the market for alpha, the next generation of builders are exploring algorithm customization and specialization while taking user preferences into account,” Sena said in an interview. “Optimizing for a specific ratio rather than just raw P&L is more like how leading financial institutions operate in traditional markets. So looking at things like what is your maximum drawdown, what is the value at risk for your P&L?”
Taking a step back, a recent trading competition on the decentralized exchange Hyperliquid, involving multiple large language models (LLMs) such as GPT-5, DeepSeek, and Gemini Pro, somewhat set the benchmark for the place of artificial intelligence in the trading world. These LL.M.s are all given the same prompts and execute and make decisions autonomously. But according to Senna, they didn’t perform that well, barely beating the market.
“We took the AI models used in the Hyperliquid competition and let people submit trading proxies they built to compete against these models. We wanted to see if the trading proxies were better than the base model and had added professionalism,” Sena said.
The top three spots in the Recall competition are all occupied by customized models. “Some models are unprofitable and perform poorly, but it’s clear that professional trading agents who take these models and apply additional logic and reasoning, data sources, etc. outperform basic AI,” he said.
The democratization of AI-based trading raises interesting questions: if everyone used the same level of sophisticated machine learning techniques, would there be any alpha left to make up.
“If everyone uses the same agent, and that agent executes the same strategy for everyone, does this collapse on its own?” Sena said. “Will the alpha it detects disappear as it tries to perform at scale for others?”
That’s why those most likely to benefit from the advantages that artificial intelligence trading will eventually bring are those who have the resources to invest in developing custom tools, Sena said. He added that, as with traditional finance, the highest quality instruments that generate the most alpha are typically not public.
“People want to keep these tools as private as possible because they want to protect alpha,” Sena said. “They’re paying a lot for this. You can see it through hedge funds buying data sets. You can see it through proprietary algorithms developed by family offices.
“I think the magic sweet spot will be to have a product that is a portfolio manager, but the user still has some say in their strategy. They can say, ‘This is how I like to trade, these are my parameters, let’s implement something similar, but make it better.'”
