Google Launches Open-Source AI Agent Runtime and Kubernetes-Based Substrate to Power Massive, Long-Running Agent Workflows

Google launches Agent Executor and Agent Substrate, two open-source tools designed to power massive, long-running AI agent workflows spanning hours or days, with Kubernetes-based infrastructure capable of scaling to hundreds of millions of agent instances.

• Google is launching Agent Executor, an open-source runtime standard designed to handle long-running AI agent workflows that can span hours or days, offering capabilities like durable execution, secure isolation, session consistency, connection recovery, and trajectory branching.
• Agent Executor bridges multiple deployment models, allowing enterprises to mix and match Google-built agents, custom managed agents, and third-party frameworks like LangChain and ADK, while maintaining full sovereignty over their own infrastructure, data, and compute environments.
• Google is also unveiling Agent Substrate, a new open-source Kubernetes-based project built to scale agent workloads to hundreds of millions of instances, introducing a specialized control plane optimized for the rapid, sub-second tool calls that overwhelm traditional Kubernetes setups.