Pearl is Olas’ consumer gateway to a future where tiny AI agents silently trade, direct and create predictive markets at a scale humans will never touch, said co-founder David Minasch.
Summary
- Olas founder David Minasch traces Pearl back to early agent work at Fetch.ai and Valory, then pivots from B2B DAO tools to a consumer app with AI agents.
- Pearl reverses strong, long-term agents like Polystrat, which filter Polymarket markets, use predictive tools and sometimes outperform human traders by 2–3x.
- Minasch sees prediction markets as the reasons for training the economics of AI, with agents already having a large share of the work and the long tail of markets that are largely offered by machines, under real control.
David Minasch sat down with crypto.news on March 31 on the sidelines of ETHCC to explain why Pearl’s long-running AI agents are remaking prediction markets from the inside out.
From Fetch.ai to Pearl
Minasch’s route to autonomous agents is a textbook crypto-AI encounter. “I was drawn to the space because of my background in economics and game theory,” he told crypto.news, recalling his move to crypto after several years working in machine learning.
At Fetch.ai, where he spent two years, his team “created the first agent framework in crypto almost,” adhering to a simple but comprehensive idea: “you can have a kind of wallets that are not controlled by humans but by machines.”
“We actually wrote a detailed paper on this, which was ahead of its time,” he adds. In 2021, he took those lessons to Valory, the core lab behind Olas, which has been testing a range of applications and go-to-market strategies.
The first bet was B2B: independent agents sold on DAOs like CowSwap, Balancer and Ceramic. “It went well but it never really happened,” Minasch admitted. The real pivot came in 2023, when “general purpose major language models used like ChatGPT” arrived and Olas “shifted more to B2C.” Pearl is the result: a “multi-agent B2C system,” built for users, not management platforms.
By the time Pearl launched in February 2025, the entire industry had grasped Olas’ first agent thesis. “The crypto space and the AI space have come close to agents, now everyone is building agents or using agents or both,” Minasch said. But he argues that many people’s perception of an agent is still being shaped by social media platforms like ChatGPT: “the interactive driver experience” where you tell and it responds, in front of you, in real time.
Olas is clearly betting against that dominant pattern. “When you have long-term agents who have the same independence but are strong so they can’t do anything but do interesting things within a certain area. That’s when it’s really fun,” he said. Pearl is designed to be around those with tighter scope, background processes rather than being general assistants, Minasch points out.
“With Pearl, we are deliberately reducing the number of agents,” he explained. He points to new tools like OpenClaw—such as verification and warning. “OpenClaw confirmed many of our core ideas that people want a first-hand experience with AI agents,” he says, but “the product can do too much, which causes a lot of problems, including security, but also a problem for the user.”
In his view, that kind of system is designed for tinkerers who “just want to shape this thing into something useful for them.” The “low friction user” wants to “just press a button” and get a consistent result. “I have it and I’ve asked it to send me a daily report and a broken part of the time,” he said of OpenClaw. “That’s not a good product.” Pearl’s agents, in contrast, are designed to do one thing—trading, yield seeking, market making—reliably. Limited range, high definition, low latency.
Polystrat is a pure expression of that philosophy. “Polystrat is an example because here’s just an idea: give some money, trade it in predictive markets,” Minasch said. Instead of dealing with Polymarket’s UX—wallet setup, funding, market selection, position sizing—the user sends funds to Polystrat and lets the agent do the work.
“Polystrat is just like a Polymarket user,” he insists. “If you want to use Polymarket you as a person need to set up a wallet, fund it and then face the decision of which market to trade. Polystrat presents all this and the idea is that it just trades for you.” The agent focuses on the country and political news markets, which are “not very permanent” and usually close “within the next four to five days.”
Technically, the flow is easy but merciless. The agent filters markets using rules such as currency and expiration time, then uses “prediction tools,” which Mnasch describes as “workflows that sit on top of models and data sources.” “There are many different forecasting tools and the agent learns over time which ones to take and which ones not to take,” depending on the market. The price and size location engine converts those predictions into positions and the system automatically trades on your behalf.
Performance wise, Polystrat distinguishes between 56 and 69% accuracy, Minasch says. As ships, “our agents… Individual Polystrat cases, however, can bring “up to 100% ROI in total and as much as 100% ROI per trade.” The goal is not anecdotes but the edge of statistics: “having Polystrat ships on average is a good ROI.”
Trade is only part of the story. As more agents enter Polymarket and its predecessors, Minasch sees prediction markets becoming “early examples of these market-driven AI systems… places that incorporate truth discovery at economic scale.”
He doesn’t pretend the instruments are clean. In controversial questions—or markets with conflicting results—late information and conflicting results are common. Polystrat or other agents at Pearl try to solve that. “Polystrat itself is a commercial agent over Polymarket,” not a consensus building or truth serum.
But AI is already reshaping participation, creation and policing. “It is not clear how many traders in the prediction markets are already AI agents but it is probably more than 30%,” believes Minasch. “Probably more than half already,” he adds. Therefore, people have a limited attention span, so “the entire long tail of the prediction markets will be given to AI agents,” he predicts.
Most importantly, Minasch breaks with crypto libertarianism in governance. “We take the view that there should be regulation of prediction markets,” he said, citing markets that “look like murder markets” or that “encourage immoral behavior.” “With some degree of regulation or self-regulation,” more markets and more AI participants should “drive prices to balance” and “enhance embedded information in the markets,” opening the door to derivatives, hedging and other built-in tools.
Asked whether Olas agents could be “automated data providers across multiple networks,” Minasch brushes aside the distinction. “Liquidity provision is an effective and trading strategy,” he says.
In that framework, Pearl is not a single application and it is an application for small, long-term agents: Polystrat for speculative markets, Optimus for yield, Omenstrat for market creation and whatever follows to get money available in all areas. A consistent design choice is scope: each agent does one thing, at remote locations, with as little human intervention as possible.
“We were just starting something that a lot of people are doing now,” Minasch said of the agent wave. The difference now is that Pearl is pushing those agents into retail products, turning speculative markets into a playground and proof-of-concept for AI-driven capitalization and discovery.
