Olam Labs evaluates AI agents in multi-agent simulations to measure social intelligence, deception, negotiation, cooperation, and real-world agentic performance.