AI agents are quietly rewriting prediction market trading

Prediction markets have long promised to aggregate insights about future events. Increasingly, those signals are coming not just from people, but from machines.

According to David Minarsch, CEO and co-founder of Valory AG, the team behind the crypto-AI protocol Olas, autonomous AI agents are emerging as powerful tools for trading prediction markets, particularly for retail users trying to compete in an increasingly automated environment.

Valory builds products at the intersection of blockchain and multi-agent systems (MAS), and its current focus is Olas, formerly known as Autonolas. The protocol is designed as infrastructure for autonomous software agents that can run services on blockchains, interact with smart contracts, and cooperate with one another while earning crypto rewards.

The broader vision is what Minarsch calls an “agent economy”. A decentralized ecosystem where autonomous AI agents perform useful tasks and generate value for their users.

One of the most visible experiments in that vision is Polystrat, an AI agent launched on the prediction-market platform Polymarket in February 2026. The agent trades on behalf of users who self-custody and own it, executing strategies continuously around the clock.

“In a nutshell, Polystrat is an autonomous AI agent that trades on Polymarket 24/7 on behalf of its human user,” Minarsch said. The idea is simple: while humans sleep, work or lose focus, the agent keeps trading.

Prediction markets, platforms where users trade contracts tied to real-world outcomes, have surged from niche forecasting tools into a fast-growing corner of fintech over the past few years. The industry’s breakout moment came during the 2024 U.S. presidential election, when trading volumes spiked and the markets gained mainstream visibility, followed by rapid expansion into sports, economics, and crypto-related bets. By 2025, total notional trading volume across major platforms exceeded $44 billion, with monthly activity reaching as much as $13 billion during peak periods.

Today the market is highly concentrated around two dominant players: Kalshi, a U.S.-regulated event-contracts exchange overseen by the Commodity Futures Trading Commission, and Polymarket, a crypto-native platform that operates globally and offers a broader range of prediction markets. Together they account for roughly 85–97% of trading volume in the sector, processing tens of billions of dollars in annual bets on everything from elections and central-bank policy to sports and cultural events

Why machines may outperform humans

The push toward AI-driven trading stems from a simple observation. Much of the intelligence embedded in modern AI models hasn’t yet translated into financial markets.

That realization prompted Valory’s team to begin building what they call a “prediction market economy” on Olas in 2023, an ecosystem where AI agents use prediction tools and data pipelines to forecast outcomes and trade on them.

Prediction markets themselves are built on probabilistic forecasting. A simple guess about an event, whether a political outcome, economic indicator or sports result, might be no better than a coin flip. But structured data analysis and disciplined trading strategies can change that equation.

“Simply prompting off-the-shelf models with markets usually results in outcomes no better than a coin-flip,” Minarsch said. “But state-of-the-art AI models wrapped in custom workflows, so called prediction tools, have historically shown predictive accuracy up to 70% and higher.”

The results so far suggest that machines may have an advantage. Third-party data indicates that only about 7% to 13% of human traders achieve positive performance on prediction markets, while the majority lose money.

At the same time, machine participation is growing quickly. More than 30% of wallets on Polymarket are already using AI agents, according to analytics platform LayerHub.

Minarsch believes this trend reflects a broader shift: humans are already competing with machines whether they realize it or not. “You have human participants in prediction markets alongside many machines,” he said. “So humans are already in a battle with machines.”

The key difference is that machines are less emotional and better at sticking to consistent strategies.

By making AI agents available to everyday users, Olas aims to level that playing field.

Early traction for autonomous traders

The early performance of Polystrat has been encouraging.

Within roughly a month of launch, the agent executed more than 4,200 trades on Polymarket and recorded single-trade returns as high as 376%, according to data shared by the team.

“Agents tend to do better than humans,” he said. “Polystrat AI agents already outperform human participants in Polymarket, with over 37% of them showing a positive P&L versus less than half that number for human participants.”

Users can configure their own agents depending on strategy preferences, data sources or risk tolerance.

The long tail of prediction markets

Beyond performance, Minarsch believes AI agents could unlock an overlooked opportunity in prediction markets: the “long tail” of niche or localized questions.

Many prediction markets revolve around major global events, elections, macroeconomic data or high-profile sports competitions. But countless smaller questions remain largely unexplored.

“Humans often don’t bother digging for the information,” Minarsch said. “They can’t be bothered to make the effort.” AI agents, by contrast, can analyze large numbers of smaller markets simultaneously.

“The long tail of prediction markets is very interesting for AI agents,” he said. “You just point the agent at the problem and it does the work.”

This could help expand prediction markets as a data-gathering tool for businesses, policymakers and decision-makers. Forecast markets have long been studied as ways to aggregate dispersed knowledge and surface insights that traditional surveys or models might miss.

In that sense, prediction markets may become a kind of upstream technology for decision-making across industries.

Human-AI collaboration

Despite the rise of automation, Minarsch does not see AI agents replacing humans entirely.

Instead, he frames them as complements.

“Humans make choices in a more rushed way, which can be detrimental,” he said. “AI agents can act as something humans rely upon.”

One future direction involves allowing users to augment their agents with proprietary knowledge or specialized data sets. “We see demand from users who want their agent to tap into their own knowledge base or proprietary information,” Minarsch said. “That would allow agents to trade in a more principled way than a human could.”

Over time, the team says prediction models and data pipelines powering these agents have improved significantly, generating sustained alpha when combined with general-purpose large language models.

Risks and regulation

The growth of prediction markets also raises ethical and regulatory questions.

Some critics argue that markets forecasting wars, deaths or disasters could create incentives to manipulate outcomes or profit from harmful events.

Minarsch acknowledged that careful guardrails are needed.

“There needs to be regulation about what kinds of prediction markets should exist,” he said.

At the same time, he believes AI agents could also help detect problematic markets or manipulation attempts by identifying suspicious patterns.

“Agents could spot patterns and help shut down problematic markets,” he said.

Building a user-owned AI economy

For Minarsch, the ultimate goal is not simply better trading strategies.

It is ensuring that everyday users retain a stake in an increasingly automated digital economy.

A future where AI systems perform most economic activity could risk disenfranchising individuals if centralized platforms control the technology. “Olas aims to create a world where human users can be empowered through their AI agents rather than disenfranchised by them.”

To counter that dynamic, the project emphasizes user ownership of AI systems. “We want to create more user-owned agents,” Minarsch said.

If successful, that model could allow people to deploy autonomous software that generates value on their behalf across markets and services. Prediction markets are just the starting point.

Read more: AI rout hits software stocks, but Grayscale says blockchains stand to benefit