Google's Open Agentic Orchestrator

Running AI agents becomes more manageable and cost-effective as Google's new AX orchestrator allows for efficient execution of billions of tasks with quick suspend and resume.

Google has open-sourced AX, a new agentic orchestrator designed for running AI agents at scale. AX provides declarative primitives for defining tasks, managing isolated workspaces, controlling network access, and configuring models. It aims to simplify the development and deployment of agentic workloads, which are characterized by their stateful, bursty, and long-running nature, often involving interactions with model APIs and tool servers. The system is built on Google's Agent Substrate, enabling it to handle billions of concurrent agent sessions per cluster with sub-second suspend and resume capabilities. AX allows developers to define agent requirements, such as Git repositories and desired environments, using a YAML configuration. It also integrates generative AI features, enabling environment setup through plain English descriptions. The orchestrator is designed for rapid iteration, supporting various agentic applications from interactive coding agents to large-scale research experiments.

AX introduces a novel approach to orchestrating AI agents, treating them as a distinct workload type that differs from traditional microservices or batch jobs. Its declarative syntax and focus on primitives like Task, Workspace, Gateway, and Model abstract away complex infrastructure management. This allows developers to focus on agent logic rather than the underlying compute, networking, and state persistence. The platform's ability to multiplex dozens of tasks onto shared worker resources and achieve sub-second resumption is key to its cost-effectiveness and performance for AI workloads that are often idle while awaiting model or tool responses.

The open-sourcing of AX, under an Apache 2.0 license, signals a move towards standardization and broader adoption of specialized agentic infrastructure. By building on Agent Substrate, AX leverages Google's extensive research and operational experience in large-scale isolation and scheduling. This release aims to lower the barrier to entry for researchers and developers building complex agentic systems, potentially accelerating innovation in areas like autonomous agents, large-scale experimentation, and interactive AI applications. It provides a production-ready runtime for these novel workloads, contrasting with the experimental nature of many current agent development frameworks.