Race Condition

In collaboration with Google, we created Race Condition for the Google Cloud Next 2026 Developer Keynote. A world-first, open-source 3D multi-agent simulation that brings the complexity of a city-scale marathon through Las Vegas to life. Presented live on stage, the experience followed autonomous agents as they navigated and responded to real-world conditions and made race decisions in real time.

The experience is a dynamic simulation environment where AI agents can navigate the city, respond to changing conditions and make race decisions in real time. By recreating the marathon route as a living system, we gave the Race Condition experience a world that agents could reason within, adapt to and bring to life.

One of the project's most significant UX breakthroughs is the deployment of the Agent-to-User Interface (A2UI) Protocol. Moving past traditional, text-heavy LLM outputs, the autonomous AI agents generate a structured schema. The Angular frontend intercepts this and dynamically renders native, fully interactive UI on the fly, fundamentally redefining human-in-the-loop digital experiences.

Each of the 1,000+ runners operates as an independent autonomous agent generating its own contextual "thoughts," pacing strategies, and real-time reactions. To process these thousands of concurrent LLM inference loops effortlessly, the architecture utilizes Google Kubernetes Engine (GKE) to dynamically provision fleets of compute containers running open-weight Gemma models. This setup serves to demonstrate how enterprise cloud infrastructure can handle high-density, real-time AI workloads at a massive scale.

By open-sourcing Race Condition, we helped turn the keynote experience into a reusable foundation for others to explore, build on and learn from, extending the impact beyond the stage.