AI-based trading has not yet reached its “iPhone moment,” where everyone carries a reinforcement-learning algorithmic portfolio manager in their pocket, but something like it is coming, experts say.
In fact, the power of AI finds itself confronting the dynamic, conflicting arena of trading markets. Unlike an AI agent that is informed by endless loops of self-driving cars that learn how to accurately recognize traffic lights, no amount of data and modeling will be able to predict the future.
This makes improving AI-based trading models a complex and demanding process. The measure of success is usually a profit and loss (P&L) measurement. But advances in how algorithms are customized are creating agents that are constantly learning how to balance risk and reward when faced with a variety of market conditions.
Allow risk-adjusted metrics e.g Sharpe ratio Michael Sina, the company’s chief marketing officer, said that enriching the learning process doubles the complexity of the test Call the laboratoriesa company that runs 20 or so AI trading arenas, where the community introduces AI trading agents, and these agents compete over the course of four or five days.
“When it comes to scanning the market for alpha, the next generation of creators is exploring algorithm customization and specialization, keeping user preferences in mind,” Sina said in an interview. “Optimizing for a certain percentage and not just the raw P&L is more similar to the way leading financial institutions operate in traditional markets. So, given things like, what is the maximum withdrawal, what is the amount of value at risk to achieve that P&L?”
take a step back, Recent trading competition on decentralized exchange Hyperliquidwhich includes several large language models (LLMs), such as GPT-5, DeepSeek, and Gemini Pro, kind of sets the baseline for where AI will be in the world of commerce. All of these MBAs were given the same direction and carried out independently, making decisions. But it wasn’t that good, according to Senna, and it barely beat the market.
“We took the AI models used in the Hyperliquid competition and allowed people to submit their own trading agents they had created to compete against those models. We wanted to see if the trading agents were better than the basic models, with this added specialization,” Sina said.
The top three spots in the Recall competition were taken by custom models. “Some of the models were unprofitable and underperformed, but it became clear that specialized trading agents that take those models and apply additional logic and heuristics and data sources and things on top are outperforming basic AI,” he said.
Democratizing commerce based on artificial intelligence Raises interesting questions About whether there would be any alpha left to cover if everyone was using the same level of sophisticated machine learning technology.
“If everyone is using the same agent and that agent is executing the same strategy for everyone, does that kind of collapse itself?” Sina said. “Does the alpha he discovers disappear because he tries to implement it broadly for everyone?”
This is why those who are best placed to take advantage of the advantage that AI trading will eventually bring are those with the resources to invest in developing custom tools, Sina said. As in traditional finance, the high-quality instruments that generate the most alpha are not public, he added.
“People want to keep these tools as private as possible, because they want to protect the alpha,” Sina said. “They paid a lot for it. You saw that with hedge funds buying datasets. You can see that with proprietary algorithms developed by family offices.”
“I think the sweet spot would be when there is a product that acts as a portfolio manager but the user still has a say in their strategy. They can say: ‘This is how I like to trade and these are my parameters, let’s do something similar, but make it better.’”
