David Minarsch, co-founder of Olas, told crypto.news on the sidelines of ETHCC that his consumer app Pearl is reshaping prediction markets from within using narrow, long-running autonomous agents. According to Minarsch, one such agent—Polystrat—trades on Polymarket with an accuracy rate of 56-69%, and the fleet overall has performed 2-3 times better than human traders.
From Fetch.ai to Pearl: The B2B-to-B2C Pivot
Minarsch's journey began with a background in economics and machine learning. He spent two years at Fetch.ai building one of crypto's earliest agent frameworks, centered on the idea of wallets controlled by machines. In 2021 he co-founded Valory, Olas's core lab, initially targeting DAOs like CowSwap and Balancer with B2B autonomous agents. "That went okay but never really took off," he admits. The turning point came in 2023 with the rise of general-purpose LLMs like ChatGPT, prompting Olas to pivot to B2C. Pearl is the result: a consumer app housing multiple agents.
Why Narrow, Long-Running Agents Win
By the time Pearl launched in February 2025, the industry had caught up. Minarsch argues most people still envision agents as chat interfaces — a synchronous co-pilot where you prompt and it replies instantly. Olas bet against that. "With Pearl we intentionally go very narrow in terms of capabilities but allow agents to run over long horizons." He contrasts OpenClaw, which validates users' desire for local-first AI but suffers from feature bloat and security issues. Pearl's agents are designed to do one thing reliably — trade, yield farm, create markets — with a single button press.
Polystrat: Automated Trading on Polymarket
Polystrat exemplifies this philosophy. It acts as a Polymarket user: filtering markets by liquidity and time to close, applying prediction tools to generate signals, and executing trades through a local pricing and sizing engine. Users simply deposit funds and the agent handles the rest. Polystrat focuses on geopolitical and political news markets with 4-5 day durations. Individual instances have delivered 100%+ ROI, though the fleet isn't yet break-even overall. The goal is statistical edge: a fleet-wide positive average ROI.
Prediction Markets as AI Training Grounds
Minarsch sees prediction markets becoming "early prototypes for market-driven AI systems" as more agents enter platforms like Polymarket. He acknowledges disputes and information lags on contested questions, but notes AI is already reshaping participation, creation, and oversight. "It's unclear exactly how many traders are already AI agents, but it's probably over 30% — potentially more than half." He predicts the "long tail of prediction markets will be served by AI agents" because human attention is limited. Crucially, he breaks from crypto libertarianism: "We take the view that there should be regulation of prediction markets" to prevent assassination markets or incentives for bad behavior. With proper regulation, more AI participants should "drive prices to equilibrium" and improve information embedded in markets.
Pearl's Long-Term Vision: Data Liquidity Providers
Asked whether Olas agents could become autonomous "data liquidity providers" across multiple networks, Minarsch says liquidity provision is effectively a trading strategy. In that framing, Pearl is less a single app and more an operating system for narrow, long-running agents: Polystrat for prediction markets, Optimus for yield, Omenstrat for market creation. Each agent does one thing over long horizons with minimal human intervention. "We were just very early to something that a lot of people are now doing," Minarsch says. Pearl is pushing these agents into consumer products, turning prediction markets into both a playground and a proving ground for AI-driven liquidity and truth discovery.