The landscape of AI-powered trading is evolving, yet it hasn’t reached a level akin to the “iPhone moment”—the day everyone carries an algorithmic trading assistant in their pocket. Experts predict, however, that such a transformation is on the horizon.
AI’s capabilities face unique challenges in the fast-paced and competitive world of trading markets. Unlike AI models trained on the repeatable processes of self-driving cars, the unpredictable nature of financial markets makes forecasting inherently difficult, regardless of the data available.
This complexity makes the refinement of AI trading models a demanding endeavor. Traditionally, success has been measured through profit and loss (P&L). However, innovations in algorithm customization are enabling the creation of agents that learn to balance risk and reward amidst various market conditions.
Incorporating risk-adjusted metrics, such as the Sharpe Ratio, into the learning process enhances the sophistication of trading models. Michael Sena, CMO at Recall Labs, noted that their AI trading arenas have showcased a variety of AI agents competing over several days.
“The new wave of developers in the trading space is delving into algorithm customization, factoring in user preferences,” Sena elaborated. “Optimizing for specific ratios rather than just P&L resembles the operations of top financial institutions in conventional markets. One must consider parameters like maximum drawdown and value at risk to achieve that kind of P&L.”
Additionally, a recent trading competition held on the decentralized exchange Hyperliquid featured numerous large language models (LLMs) like GPT-5, DeepSeek, and Gemini Pro. They executed trades autonomously based on identical prompts, but their performance was underwhelming, barely surpassing market averages, according to Sena.
“We took the AI models from the Hyperliquid contest and allowed individuals to submit their custom trading agents to challenge those models. Our goal was to assess if these specialized trading agents could outperform the foundational ones,” Sena explained.
The results were clear, with customized models securing the top three positions in Recall’s competition. While some agents performed poorly, the evidence indicated that specialized trading agents, leveraging additional logic and data sources atop foundational models, were more successful in achieving superior results.
The rise of AI-driven trading opens up intriguing questions about the future of market alpha. If everyone employs identical advanced machine-learning technologies, will alpha dissipate?
“If a uniform agent executes the same strategies for all users, does that strategy lose its effectiveness due to overcrowding?” posed Sena. “Will the alpha disappear as it’s replicated at scale?”
Therefore, those with the capacity to cultivate custom AI trading tools are likely to reap the benefits of this evolving landscape, according to Sena. He emphasizes that, similar to traditional finance, the most effective tools that generate the highest alpha typically remain proprietary.
“The desire to keep these tools confidential stems from the need to protect valuable alpha,” Sena stated. “Financial entities invest heavily in these resources, much like hedge funds purchasing exclusive datasets and family offices developing proprietary algorithms.”
“The true breakthrough will be a product that acts as a portfolio manager while still allowing users to customize their strategies. They should be able to articulate their trading preferences and parameters, leading to smarter implementations.”
