How to design a multi-agent system that skips the LLM

A technical deep-dive into the Race Condition project: a 1000-agent marathon system built on Google ADK that makes zero LLM calls by using deterministic code where appropriate. Casey West demonstrates when to use LLMs versus deterministic logic, specifically for NP-hard route planning problems. The video teaches AI engineers how to architect multi-agent systems that route decisions to code instead of the LLM, scaling to thousands of stateless sessions.