Pearl: Redefining Prediction Markets with AI

Pearl serves as Olas’s consumer portal to a future where specialized AI agents efficiently trade, curate, and create prediction markets on a scale beyond human capabilities, according to co-founder David Minasch. In a discussion with crypto.news on March 31 at ETHCC, Minasch elaborated on how Pearl’s narrow-function AI agents are transforming prediction markets internally.

Key Takeaways

Pearl is an AI-driven platform that utilizes specialized agents to trade and create prediction markets efficiently.
David Minasch, co-founder of Olas, believes these long-functioning agents can operate autonomously without human management.
The platform emphasizes a business-to-consumer approach, contrary to traditional B2B strategies.
Pearl's agents are designed for specific tasks, ensuring reliability and reducing complexity for users.
Polystrat is a key example of efficient market trading, showing significant returns on investment with minimal user involvement.
The rise of AI agents in prediction markets challenges traditional human trader participation while highlighting the need for regulatory oversight.

Pearl AI agents trading on Polymarket dashboard

Transition from Fetch.ai to Pearl

Minasch's journey into autonomous agents epitomizes the fusion of cryptocurrency and AI. "My interest in the space stemmed from my background in economics and game theory," he shared, recalling his journey into crypto after several years of working on machine learning applications.

At Fetch.ai, where he spent two years, his team developed potentially the first agent framework in the crypto space, based on a significant yet simple idea: allowing machines, rather than humans, to control wallets.

"We actually wrote a detailed paper on this, which was groundbreaking," he notes. In 2021, he transitioned those insights into Valory, the foundational lab behind Olas, which has since experimented with various applications and market strategies.

The initial focus was on business-to-business (B2B) offerings, selling autonomous agents to decentralized autonomous organizations (DAOs) like CowSwap, Balancer, and Ceramic. "That went reasonably well but never really took off," Minasch admits. The substantial shift occurred in 2023, coinciding with the arrival of “general-purpose large language models like ChatGPT,” prompting Olas to pivot more toward business-to-consumer (B2C) applications. This shift led to the creation of Pearl, which focuses on end users rather than governance forums.

By the time Pearl was launched in February 2025, the entire industry had progressed toward Olas’s early vision of agents. "Both the crypto and AI sectors shifted towards agents; now, everyone is either building agents or utilizing them," Minasch states. However, he believes that many people's understanding of an agent still revolves around chat interfaces like ChatGPT, which provide a "co-pilot synchronous experience" where prompts and responses occur in real-time.

Olas is opting for an alternative approach. "When you have long-running agents that maintain autonomy while being tightly scoped, they can carry out specific interesting tasks," he explains. Pearl is built around these tightly constrained background processes instead of general-purpose assistants, as Minasch emphasizes.

“With Pearl, we aim deliberately to narrow the capabilities of an agent,” he clarifies. He references tools like OpenClaw as both a validation and a cautionary tale. “OpenClaw confirmed many of our assumptions that users seek local-first experiences with AI agents,” he says, but "the product can be overly complex, leading to issues such as security concerns and user difficulties."

In his opinion, that type of system caters to tinkerers who prefer to customize it to their needs. In contrast, "low-friction users" desire simplicity; they want to "just press a button" and see consistent results. Minasch illustrates this with his experience using OpenClaw: "I requested a daily report, but it often fails to deliver." This inconsistency results in an unsatisfactory user experience. Pearl’s agents, on the other hand, are focused on mastering singular tasks—such as trading, yield-seeking, and market creation—ensuring reliability through their limited scope.

Creating Synergy with Polystrat

Polystrat exemplifies this focused philosophy. "Polystrat represents the concept of providing capital and allowing it to trade in prediction markets," Minasch mentions. Instead of grappling with Polymarket's user experience involving wallet setups, funding, market selections, and position sizing, users delegate their funds to Polystrat, letting the agent handle the complexities.

“Polystrat behaves like a user of Polymarket,” he emphasizes. “If you want to engage with Polymarket, you, as a human, must navigate the setup and trading decisions. Polystrat takes care of that, trading on your behalf.” The agent primarily engages with geopolitical and political news markets, which often conclude within four to five days.

The architecture of the system is straightforward yet exacting. The agent evaluates markets based on criteria like liquidity and closure time, followed by applying "prediction tools," defined by Minasch as “workflows that operate atop models and data sources.” There are various predictive tools available, and the agent learns over time which to employ based on market conditions. A local pricing and sizing engine processes these predictions into trading positions, allowing the system to trade autonomously.

In terms of performance, Polystrat achieves an accuracy rate of between 56% and 69%, according to Minasch. "As a group, our agents have outperformed human traders by two to three times," although they have not yet reached a break-even point overall. Individual Polystrat instances can reportedly yield as much as a 100% return on investment (ROI) and multiple 100% returns on individual trades. The primary aim is to establish a statistical advantage so that the average ROI for the Polystrat fleet remains positive.

However, trading isn’t the whole narrative. As more agents participate in Polymarket and its predecessors, Minasch expects prediction markets to evolve into "early prototypes for market-driven AI systems—platforms that integrate truth discovery at economic scales."

He is realistic about the challenges. In markets with contentious questions or outcomes, delays and disputes are frequent. Neither Polystrat nor other agents on Pearl aim to resolve these issues. "Polystrat itself is merely a trading agent functioning atop Polymarket; it doesn’t serve as a consensus builder or a truth-telling mechanism."

AI is already transforming how participants engage, create, and regulate. "It’s uncertain precisely how many traders in prediction markets are AI agents," Minasch estimates, “but likely over 30%, perhaps even more than half.” Since humans face limited attention spans, he predicts that "the long tail of prediction markets will largely be serviced by AI agents."

Crucially, Minasch diverges from the typical crypto libertarian stance regarding governance. "We believe that prediction markets should be regulated," he asserts, referencing markets that resemble "assassination markets" or those "promoting harmful behaviors." With “a modest amount of regulation or self-regulation,” increased market activity and participation by AI could "drive prices toward equilibrium" and "enhance the quality of information in markets," paving the way for derivatives, hedging, and additional financial instruments.

When asked if Olas agents could evolve into "data liquidity providers autonomously operating across various networks," Minasch dismisses the distinction. "Liquidity provision is effectively another form of trading strategy," he contends.

From this perspective, Pearl is less a single application and more of an operating system for specialized, long-running agents: Polystrat targeting prediction markets, Optimus focused on yield, Omenstrat for market creation, and whatever follows for liquidity across platforms. The consistent design philosophy remains the same: each agent specializes in one task over extended periods with minimal human intervention.

“We were pioneers in something that many are now adopting,” Minasch reflects on the emergent agent trend. The current reality is that Pearl is advancing these agents into consumer-facing products, transforming prediction markets into both playgrounds and proving grounds for AI-driven liquidity and truth discovery.