AI For Prediction Markets

David Enem explains agentic AI, PolyStrat trading agents, and building user-owned agent economies with practical safety and commerce lessons.

Key Takeaways

PolyStrat runs in the Pearl desktop app—fund by card or crypto, start autonomous trading, then customize strategies; design intentionally limited for safety.
Agents trade by fetching public data, prompting LLMs for outcome probabilities, then applying rule-based trade sizing and selective market choice—no private data required.
OLAS protocol enables agent-to-agent commerce and micropayments; the marketplace supports agent intelligence exchange and has processed 14M+ on-chain transactions.
Safety-first design: constrain agent permissions to prevent key leakage, illegal actions, or prompt-injection exploits; isolate harmful agents and acknowledge operator liability.
Social and crypto outlook: expect agent-focused social subnets and agent wallets to emerge by 2026, with tension between user-owned co-ops and centralized AI lab control.
Product evolution: moved from DAO/B2B to user-facing, off-chain agent products after LLM APIs; Pearl is open-source, self-custodied, free to use but requires OLAS staking.