Two Google developer advocates discuss Race Condition, an open-source large-scale agentic simulation framework built for Google Cloud Next's developer keynote. The system simulates planning a marathon in Las Vegas using a multi-agent architecture powered by Google's Agent Development Kit (ADK), A2A protocol, Pub/Sub messaging, Google Cloud Memorystore (Redis), and Gemma 4 open-weight models running on GKE with vLLM for scalable inference. Key design patterns covered include agent identity and security via agent gateway, context management strategies to control token costs, server-authoritative game loop patterns borrowed from game development, deterministic vs. LLM-powered agents for scale, and real-time state broadcasting via Pub/Sub. The project is open source and includes free Google Cloud credits for developers who fork the repository.
Sort: