John Forfar | Dev Rel

AI Agents are taking over Prediction Markets! Here's my latest walk through on Mirofish AI Agent swarm intelligence. I got 4 agents running locally, predicting the global oil price based on next attack. This same software stack can scale up to 10,000+ with enough GPU's ! YouTube link below:

I spent a few days studying the MiroFish swarm-intelligence engine Then I tried to build my own version for Polymarket. The results were insane. Instead of predicting prices, the engine simulates how people react to events. Core idea: Prediction markets don’t move on math alone. They move on behavior. So I built a multi-agent simulation. Each agent has memory, bias, and strategy. Core components: Swarm agents 168 agents / 6 clusters / 700+ links Belief update model Pₜ₊₁ = f(Pₜ, news, sentiment, liquidity) Consensus probability P_swarm = Σ wᵢ Pᵢ Market edge EV = P_swarm − P_market Monte-Carlo validation 10,000 simulations per event Network clustering detects narrative shifts before price moves The system simulates: • voters • media • donors • traders • pollsters • momentum bots • narrative spread Each agent updates beliefs → cluster reaches consensus → swarm outputs probability. If swarm probability ≠ market price → trade Current dashboard: 1,250 active agents 100,000+ interactions 3,800 graph nodes 240 simulation rounds Top prediction engine: Polymarket price: 20% Swarm probability: 27.3% Alpha edge: +7.3% Win rate: 67% Sharpe: 8+ Max DD: <2% This changed how we can see prediction markets. It’s not trading. It’s modeling reality.