Pearl: The Future of Prediction Markets and AI Liquidity

Pearl serves as Olas’s consumer interface to a future where specialized AI agents will automatically trade, curate, and create prediction markets on a scale beyond human reach, according to co-founder David Minasch. In an interview with crypto.news on March 31 during ETHCC, Minasch shared insights into how Pearl's dedicated AI agents are fundamentally transforming prediction markets.

Key Takeaways

AI Integration: Pearl represents a significant step toward leveraging narrow AI agents in trading and prediction markets.
Evolution from B2B to B2C: Olas shifted its focus from business-to-business applications to consumer-facing tools, leading to the creation of Pearl.
Focused Functionality: Pearl's agents are designed for specific tasks, such as trading and market creation, rather than general-purpose assistants.
Improved Trading Experience: The Polystrat agent simplifies the prediction market trading process for users by automating trade decisions based on market conditions.
Market Regulation: Minasch advocates for a regulatory framework in prediction markets to enhance information quality and prevent harmful practices.

Pearl AI agents trading on Polymarket dashboard

The Journey from Fetch.ai to Pearl

David Minasch's path into the world of autonomous agents reflects the merging of cryptocurrency and AI. "I was lured into this space due to my background in economics and game theory," he explained, recalling how he transitioned into crypto after years of work in machine learning. At Fetch.ai, he contributed to what might be one of the first agent frameworks in crypto, based on a groundbreaking idea: having wallets managed by machines instead of humans.

"We published an extensive paper ahead of its time," he noted. In 2021, he launched Valory to apply those insights and explore various applications and market strategies. The initial focus was on B2B, providing autonomous agents to DAOs like CowSwap, Balancer, and Ceramic. While the effort was somewhat successful, it didn’t fully take off. The real turning point was in 2023, with the emergence of widely usable AI models like ChatGPT, prompting a shift to a business-to-consumer model. Thus, Pearl was born, catering directly to users rather than governance forums.

By the time Pearl debuted in February 2025, the industry had begun to embrace the concept of agents. "Now, everyone's either building agents or utilizing them," Minasch commented. However, he believes many still associate agents with chat interfaces like ChatGPT, viewing them as synchronous assistants responding in real-time.

Olas is making a distinct wager against that prevailing notion. "We want to create long-running agents that are autonomous yet limited in scope to perform interesting tasks," he explained. Pearl's design prioritizes these niche, automated processes instead of typical generalist assistants.

Focusing Capabilities with Pearl

"With Pearl, we intentionally narrow the functionality of our agents," Minasch emphasized. He cited tools like OpenClaw as both validation and caution. "OpenClaw confirmed that users prefer localized interactions with AI agents," he said, yet it also revealed that providing too many features can lead to security issues and user frustration.

In his view, less complex systems cater to tinkers wanting to customize; the ideal "low friction user" simply wants a reliable outcome from pressing a button. "I set up OpenClaw to send me daily reports, and it only worked half the time," he noted, which led to dissatisfaction. In contrast, Pearl's agents are crafted to focus on one reliable function—whether that's trading, yield generation, or market creation—ensuring high performance with minimal issues.

Introducing Polystrat

Polystrat exemplifies this philosophy. The concept is simple: provide capital and let it trade in prediction markets. Minasch stressed the ease of using Polystrat compared to Polymarket's user experience, which requires setting up wallets, funding, and market selection. Polystrat abstracts these complexities, allowing the agent to trade on the user’s behalf.

"Polystrat acts just like any user on Polymarket," he reiterated. It zeroes in on various geopolitical and political news markets that are not overly transient, typically concluding within a few days.

On a technical level, the agent employs strict filtering criteria, such as market liquidity and time until closing, and utilizes "prediction tools"—methodologies built on diverse models and data sources. "The agent learns which tools to apply to specific markets over time," Minasch explained, streamlining the trading flow and decisions.

Polystrat's performance boasts an accuracy range of 56% to 69%. As a whole, "our agents are performing two to three times better than human traders" even though they have yet to reach a fleet-wide break-even mark. However, some individual Polystrat instances have achieved returns of up to 100%, aiming for a statistical advantage to consistently generate positive ROI.

Beyond Trading

Trading is just part of the larger picture. As AI agents continue to populate Polymarket and similar platforms, Minasch envisions prediction markets evolving into "early prototypes for market-driven AI systems, facilitating truth discovery on an economic scale."

Admitting that operational challenges exist, he noted that prediction markets often struggle with information delays and disputed results on controversial issues. Neither Polystrat nor any other agents on Pearl address these challenges directly. "Polystrat is merely a trading agent for Polymarket," he clarified; it doesn't serve as a consensus-building tool.

However, AI is already enhancing how people engage with, create, and monitor these markets. "While it’s hard to quantify, I estimate that over 30% of prediction market participants are AI agents,” Minasch said, suggesting the number might be significantly higher. "Humans only have so much attention to give, so we foresee AI agents servicing the long tail of prediction markets."

Minasch also takes a stand against unchecked governance within crypto. "We advocate for some regulation around prediction markets," he stated, citing instances where markets could resemble assassination services or promote harmful actions. He believes balanced regulation could lead to more efficient markets and improved information quality, paving the way for derivatives and other financial instruments built on these platforms.

When asked if Olas agents could evolve into "data liquidity providers operating independently across different networks," Minasch blurred the lines, explaining, “Liquidity provision is essentially a trading strategy, too.”

In his analysis, Pearl operates not just as a singular application but as an ecosystem for focused, long-term agents like Polystrat for prediction markets, Optimus for yield generation, and Omenstrat for market creation, all designed to function with minimal human involvement.

"We were just ahead of the curve, but many are now recognizing the value of agents," Minasch concluded regarding the current agent movement. The distinct advancement now is Pearl’s role in making these agents accessible to retail users, transforming prediction markets into both experimental spaces and proving grounds for AI-fueled liquidity and truth discovery.