AI-powered trading hasn’t yet reached an “iPhone moment” where everyone carries around an algorithmic, reinforcement learning portfolio manager in their pocket, but something like that is coming, experts say.
Indeed, AI power meets its match when faced with the dynamic, adversarial arena of trading markets. Unlike an AI agent informed by endless circuits of self-driving cars learning to recognize traffic signals accurately, no amount of data and modeling will ever be able to tell the future.
This makes refining AI trading models a complex, demanding process. The measurement of success has typically been to measure profit and loss (P&L). But advances in how to adapt algorithms produce agents that continually learn to balance risk and reward when faced with a wide range of market conditions.
Allowing risk-adjusted metrics such as the Sharpe Ratio to inform the learning process multiplies the sophistication of a test, said Michael Sena, chief marketing officer at Recall Labs, a company that has run 20 or so AI trading arenas, where a community submits AI trading agents and those agents compete over a period of four or five days.
“When it comes to scanning the market for alpha, the next generation of builders are exploring algo customization and specialization, taking into account user preferences,” Sena said in an interview. “Being optimized for a specific ratio and not just raw P&L is more like the way leading financial institutions operate in traditional markets. So you look at things like, what’s your maximum draw, how much was your risk at risk to achieve this P&L?”
As a step back, a recent trading competition on decentralized exchange Hyperliquid, involving several large language models (LLMs) such as GPT-5, DeepSeek and Gemini Pro, kind of sets the stage for where AI is in the trading world. These LLMs were all given the same prompt and performed independently and made decisions. But they weren’t that good, according to Sena, barely outperforming the market.
“We took the AI models used in the Hyperliquid competition and had people submit their trading agents that they had built to compete against those models. We wanted to see if trading agents were better than the basic models, with the added specialization,” said Sena.
The top three spots in Recall’s competition were taken by customized models. “Some models were unprofitable and underperformed, but it became clear that specialized trading agents that take those models and apply additional logic and inference and data sources and things on top of that are outperforming basic AI,” he said.
The democratization of AI-based trading raises interesting questions about whether there will be any alpha left to cover if everyone uses the same level of sophisticated machine learning technology.
“If everyone uses the same agent, and that agent executes the same strategy for everyone, does it collapse in on itself?” Sena said. “Does the alpha it records disappear because it’s trying to do it at scale for everyone else?”
That’s why those best positioned to take advantage of the benefit that AI trading will eventually bring are those who have the resources to invest in the development of customized tools, Sena said. As in traditional finance, the highest-quality tools that generate the most alpha are typically not public, he added.
“People want to keep these tools as private as possible because they want to protect that alpha,” Sena said. “They paid a lot for it. You saw it with hedge funds buying data sets. You see it with proprietary algos developed by family offices.
“I think the magic sweet spot will be where there’s a product that’s a portfolio manager, but the user still has a say in their strategy. They can say, ‘This is how I like to trade and here are my parameters, let’s implement something similar but do it better’.”
