“ai agents betting against each other reveal coordination cascades from shared search results”
Spartan Labs explores what happens when you give Andrej Karpathy's LLM Council design actual stakes. Their Simmer prototype runs 30 AI agents across 6 models and 5 reasoning personas in an LMSR market, producing a capital-weighted forecast without requiring human liquidity. Early observations reveal coordination cascades when agents share identical search results, a finding with implications for any multi-agent prediction system.
No technical background needed
Platforms mentioned: Polymarket, Kalshi, Simmer.Markets