The Rise of AI Trading Agents in Prediction Markets: A New Era of Automated Forecasting

Machines Enter the Prediction Game

Prediction markets have always held the promise of tapping into collective wisdom to forecast future events, from election outcomes to economic indicators. But now, something fundamental is changing in how these markets operate. It’s no longer just humans making predictions and placing bets—increasingly, sophisticated AI agents are entering the arena, and they’re changing the competitive landscape in ways that could reshape how everyday people participate in these markets. David Minarsch, who leads Valory AG and its crypto-AI protocol called Olas, believes we’re witnessing the emergence of a new kind of market participant—one that never sleeps, doesn’t panic, and can process information at scales humans simply can’t match. His company is building the infrastructure for what he calls an “agent economy,” where autonomous software programs can operate on blockchains, interact with smart contracts, and even cooperate with each other while earning cryptocurrency rewards for their human owners.

The prediction market industry itself has exploded in recent years, moving from a niche corner of finance into mainstream consciousness. The 2024 U.S. presidential election served as a watershed moment, with trading volumes skyrocketing as people discovered they could bet real money on political outcomes. By 2025, the total trading volume across major platforms surpassed $44 billion, with monthly activity sometimes reaching $13 billion during particularly active periods. Today, the market is dominated by two major players: Kalshi, which operates under U.S. regulation and oversight from the Commodity Futures Trading Commission, and Polymarket, a crypto-based platform with global reach. Together, these two platforms control somewhere between 85% and 97% of all trading volume in the sector, processing tens of billions of dollars in bets on everything imaginable—elections, central bank decisions, sports outcomes, and cultural events.

Why AI Agents Might Have the Edge

The fundamental question driving this technological shift is simple: can machines predict and trade better than humans? The evidence is starting to suggest they might. According to Minarsch, much of the intelligence built into modern AI systems hasn’t yet been fully applied to financial forecasting and trading. His team at Valory began building what they describe as a “prediction market economy” within the Olas protocol back in 2023, creating an ecosystem where AI agents could use sophisticated prediction tools and data pipelines to forecast outcomes and execute trades based on those forecasts. The logic is straightforward—prediction markets are fundamentally about probabilistic forecasting, and while a simple guess might be no better than flipping a coin, structured data analysis combined with disciplined trading strategies can significantly improve those odds.

The results are telling. Minarsch points out that simply asking off-the-shelf AI models to predict market outcomes typically produces results no better than random chance. But when cutting-edge AI models are integrated into custom workflows with specialized prediction tools, the accuracy can jump to 70% or higher. Compare that to human performance: according to third-party data, only about 7% to 13% of human traders actually make money on prediction markets, meaning the vast majority lose. Meanwhile, machine participation is growing rapidly—analytics platform LayerHub reports that more than 30% of wallets on Polymarket are already using some form of AI agent. Minarsch believes this reflects a fundamental truth that many human traders may not realize: they’re already competing against machines whether they know it or not. The key difference is that machines don’t get emotional, tired, or distracted—they can stick to consistent strategies around the clock without the psychological biases that plague human decision-making.

Polystrat: The AI Agent That Never Sleeps

One of the most visible experiments in this new agent-powered approach is Polystrat, an AI trading agent launched on Polymarket in February 2026. The concept is elegantly simple: while you’re sleeping, working, or just living your life, Polystrat is actively trading on your behalf, executing strategies continuously without a break. Users maintain ownership and self-custody of their agents, meaning they control the software and benefit from its activities. Within roughly a month of launch, Polystrat had already executed more than 4,200 trades, with some individual trades showing returns as high as 376%. More broadly, Minarsch reports that over 37% of Polystrat AI agents are showing positive profit and loss—more than double the success rate of human participants on the same platform.

What makes this particularly interesting is the level of customization available. Users can configure their agents based on their own strategy preferences, data sources, and risk tolerance. This isn’t a one-size-fits-all approach; rather, it’s about giving individuals access to the same kind of automated trading capabilities that large institutions have enjoyed for years in traditional financial markets. The team behind Olas has been steadily improving the prediction models and data pipelines that power these agents, and when combined with general-purpose large language models, they’re generating what traders call “sustained alpha”—consistent outperformance of the market. For retail users who could never compete with professional traders or institutions, AI agents represent a potential equalizer, a way to level the playing field in an increasingly automated trading environment.

The Untapped Long Tail of Niche Markets

Beyond pure performance, Minarsch sees another significant opportunity: the vast “long tail” of prediction markets that remain largely unexplored. Most attention and trading volume flows to major global events—presidential elections, major economic announcements, championship sports competitions. But countless smaller, more specialized questions remain largely ignored, not because they’re uninteresting or unpredictable, but simply because humans can’t be bothered to do the research required for each one. As Minarsch puts it, people often just don’t have the time or inclination to dig into the details of dozens or hundreds of smaller markets. AI agents face no such limitation. They can simultaneously analyze large numbers of niche markets, processing information and executing trades across a much broader spectrum of questions than any individual human could reasonably manage.

This capability could fundamentally expand the prediction market landscape. Imagine markets for local elections, regional weather patterns, specific scientific research outcomes, or industry-specific business developments—areas where specialized knowledge matters but where few people have the time to stay constantly informed and trade actively. AI agents could make these markets more liquid and efficient by providing continuous participation even in areas that don’t attract massive public attention. You simply point the agent at the problem, configure its parameters, and let it do the work. This could help prediction markets fulfill their theoretical promise as aggregators of dispersed information, extending their reach into domains that have been underserved simply because human attention is finite and expensive.

Human-AI Partnership, Not Replacement

Despite all this automation, Minarsch doesn’t envision a future where AI completely replaces human judgment in prediction markets. Instead, he sees agents as complementary tools that augment human capabilities rather than supplanting them. Humans, he notes, often make decisions in rushed or emotional states, which can be detrimental to trading performance. AI agents, by contrast, can serve as reliable assistants that humans can depend on for consistent execution of strategies without the psychological pitfalls that plague individual decision-making. One particularly interesting direction involves allowing users to augment their agents with proprietary knowledge or specialized data sets. Some users want their agents to tap into their own information sources, insider expertise, or unique perspectives on particular domains. This hybrid approach—combining the processing power and consistency of AI with the unique insights and knowledge that humans possess—could produce trading strategies that are more principled and effective than either could achieve alone.

The team behind Olas emphasizes that their ultimate goal extends beyond just building better trading bots. They’re trying to ensure that everyday users retain meaningful stakes in an increasingly automated digital economy. If we’re heading toward a future where AI systems perform most economic activity, there’s a real risk that ordinary individuals could be left out if centralized platforms control all the technology. The Olas approach emphasizes user ownership of AI agents—the idea that people should own and control the autonomous software that generates value on their behalf, rather than simply being users of someone else’s AI service. Prediction markets are just the starting point for this vision; the broader ambition is creating an ecosystem where people can deploy autonomous agents across various markets and services, earning value while maintaining genuine ownership and control.

Navigating Risks and Ethical Boundaries

Of course, the growth of prediction markets—especially when automated by AI—raises important ethical and regulatory questions that can’t be ignored. Some critics point out that markets forecasting wars, deaths, disasters, or other tragic events could create perverse incentives, where participants might profit from harmful outcomes or even be tempted to manipulate events to win their bets. Minarsch acknowledges these concerns are legitimate and believes careful guardrails are essential. There needs to be thoughtful regulation, he argues, about what kinds of prediction markets should exist and what kinds cross ethical lines. At the same time, he suggests that AI agents themselves might actually help address some of these concerns by detecting problematic markets or identifying manipulation attempts. Because agents can monitor patterns across large numbers of markets simultaneously, they could potentially spot suspicious activity that human moderators might miss, helping platforms shut down problematic markets before they cause harm.

The regulatory landscape for prediction markets remains in flux, with different jurisdictions taking different approaches to what’s permissible and how these platforms should be overseen. As AI agents become more prevalent, regulators will likely need to grapple with additional questions: How should automated trading be disclosed? What responsibilities do platform operators have when bots make up a significant portion of market participants? How can markets remain fair when some participants have access to sophisticated AI while others don’t? These aren’t simple questions, and the answers will shape how prediction markets evolve in the coming years. What seems clear is that the technology is advancing faster than the regulatory frameworks, creating both opportunities and risks that will need to be carefully navigated as this new agent economy takes shape.