{"slug": "introducing-agent-executor-googles-distributed-agent-runtime", "title": "Introducing Agent Executor, Google’s distributed Agent Runtime", "summary": "Google has introduced Agent Executor, an open-source runtime standard designed to reliably manage long-running, distributed agent workflows in production. The system addresses key operational challenges by providing durable execution, secure sandbox isolation, session consistency, and connection recovery for complex agent tasks. Agent Executor is harness-agnostic and supports flexible deployment across on-premises, managed, or custom infrastructure to prevent vendor lock-in.", "body_md": "As models and harnesses improve, agents are taking on increasingly complex tasks that can run for hours or even days. But as we push agents to do more, this has surfaced a new operational problem: long-running agent workflows are fragile and incredibly hard to manage reliably and efficiently in production.\nToday, we’re introducing Agent Executor, Google’s open-source runtime standard for agent execution, resumption, and distributed deployment. Based on what we’ve learned from solving these challenges internally, we’ve built Agent Executor to have the following native capabilities:\nDurable execution: Long-running execution requires the ability to resume after outages or agentic interruptions such as human-in-the-loop (HITL) confirmations. Agent Executor provides this backend resilience automatically for any actor (e.g., an agent, agent harness, skill, tool, or sandbox) through its event log and snapshotting.\nSecure isolation: Agent Executor isolates components in secure-by-design sandboxes to prevent harmful side effects and help ensure malicious activity cannot compromise the broader service. Sandboxes are especially useful when agents generate code or handle multiple tenants or user data concurrently.\nSession consistency: In distributed agent workflows, multiple components may attempt to update shared session state at the same time. Agent Executor’s built-in single-writer architecture helps maintain consistency and reduces the risk of corruption in that state.\nConnection recovery: In long-running agentic execution, clients may disconnect for many reasons, including network outages. Agent Executor lets clients reconnect to agents and backfills responses from the last sequence seen by the client for a better user experience.\nIn this blog, we’ll share more about Agent Executor and how you can get started.\nEnterprise adoption of agents requires orchestration across deployment models. Some teams need on-prem infrastructure for proprietary workflows, performance, or compliance, while others prefer pre-built or custom managed agents for faster time-to-value. At Google I/O, we introduced a new suite of such solutions – including Antigravity 2.0 and the Managed Agents API – designed to accelerate how teams build and scale within the agentic enterprise.\nAgent Executor bridges these deployment models, letting you mix-and-match between any or all of:\nGoogle Antigravity, Gemini’s state-of-the-art agent harness\nGoogle-built frontier agents, such as the latest Deep Research agent\nCustom agents built by you and managed by Google (e.g., via the new Managed Agents in Gemini API)\nWith Agent Executor, enterprises have maximum flexibility to maintain sovereignty over workloads and keep proprietary workflows within their self-managed compute and custom sandboxes. Your internal development teams have much more flexibility over how agents are deployed and managed and you benefit from:\nPrevent vendor lock-in: Deploy your agents on your own infrastructure without being tethered to a specific provider’s model or compute environment. This allows for full control over data residency and your cost and budgetary controls.\nBring your own harness and agents: Agent Executor is designed to be harness-agnostic, allowing you to bring your own or use those made available by other vendors. It also supports agents developed with industry-standard frameworks and protocols providing a broad ecosystem of compatible agents.\nFully control execution: Agent Executor allows developers to run the entire agentic stack, including MCPs, skills, and other agents, directly on their own data plane. Developers can choose any compute with custom isolation boundaries and workload policy enforcement.\nAs agent workloads scale into the hundreds of millions and become increasingly long-running, our customers are hitting the limits of traditional compute abstractions because unlike traditional software, agents are nonlinear programs that wait for external inputs. To solve this problem, we’ve partnered with the Google Kubernetes Engine team on Agent Substrate, a new open-source project also announced today.\nAgent Substrate introduces a new level of abstraction for Kubernetes that moves agents onto and off of ready compute capacity in real-time, resulting in lower latency with higher scale and efficiency. While standard Kubernetes is optimized to handle thousands of long-running services, Agent Substrate is designed for the chatter of millions of sub-second tool calls that would otherwise overwhelm a standard control plane. Agent Substrate takes core secure runtime and snapshotting capabilities of existing sandbox infrastructure and pairs them with a minimal control plane designed to bypass some of the limitations of Kubernetes, without reinventing the rest of it. Working together, these layers enable you to:\nMaximize compute efficiency: Agent Substrate introduces a new control plane designed to handle hundreds of millions of registered agents. Together with Agent Executor, Agent Substrate can provide a foundation for today’s largest agent deployments.\nStay within the Kubernetes ecosystem: Agent Substrate is built on top of Kubernetes and allows scheduling and horizontal scaling of compute with declarative configuration.\nIn the demo below, we showcase using Agent Executor together with Agent Substrate with a sample workload.\nModels, agents, harnesses, and the infrastructure around them are all evolving faster than ever. We’re building Agent Executor in the open so we can validate the design in the hands of real developers and improve based on your feedback.\nAgent Executor is available now in preview. We invite you to explore the code, test it with your own workloads, and help shape the future of agent runtimes. Head over to our GitHub repo to get started today.", "url": "https://wpnews.pro/news/introducing-agent-executor-googles-distributed-agent-runtime", "canonical_source": "https://cloud.google.com/blog/products/ai-machine-learning/agent-executor-googles-distributed-agent-runtime/", "published_at": "2026-05-20 16:00:00+00:00", "updated_at": "2026-05-20 16:07:12.211106+00:00", "lang": "en", "topics": ["artificial-intelligence", "open-source", "developer-tools", "cloud-computing", "enterprise-software"], "entities": ["Google", "Agent Executor"], "alternates": {"html": "https://wpnews.pro/news/introducing-agent-executor-googles-distributed-agent-runtime", "markdown": "https://wpnews.pro/news/introducing-agent-executor-googles-distributed-agent-runtime.md", "text": "https://wpnews.pro/news/introducing-agent-executor-googles-distributed-agent-runtime.txt", "jsonld": "https://wpnews.pro/news/introducing-agent-executor-googles-distributed-agent-runtime.jsonld"}}