Stop building slow AI: Optimizing multi-agent systems for production

This advanced technical session from Google Cloud Next 2026 dives into building production-optimized multi-agent AI systems using the Agent Development Kit (ADK) and Agent2Agent Protocol (A2A). Developers learn to create specialized planner, simulator, and evaluator agents, with a focus on real-time evaluation techniques using Gemini 3.1 Pro and optimizing inter-agent communication via Pub/Sub, WebSockets, and protocol buffers. The talk is designed for experienced cloud developers and ML engineers building scalable AI applications who need to understand production constraints and data governance for multi-agent architectures.