- Published on
Google's Open Agentic Orchestrator
- Authors
- Name
- DeepHorus
Google has unveiled Open Agentic Orchestrator (AX), a platform for running agentic tasks at massive scale. AX treats agents as a distinct workload class, offering declarative primitives for isolated execution, workspace setup, network policy, and model configuration. It runs on Agent Substrate, a compute runtime built for high density and fast stateful actor lifecycles.
Developers define tasks in YAML, specifying Git repos, MCP servers, skills, or plain-English goals. AX sandboxes each task, wires up its workspace, fences its network, and can run billions of tasks per cluster. Teams can use one task per agent or compose many tasks as needed.
Key features include CPU and memory limits, easy workspace setup, network policy gateway, and centralized model configuration. Idle agents waiting on model responses or human input are checkpointed and suspended, then restored in under a second with zero cold-start delay. Dozens of tasks share worker resources, so users pay only when agents are actively running code.
The platform supports interactive coding agents, long-running agent servers, Jupyter notebooks, headless browser testing, and custom tool runtimes. It also offers generative workspaces, where an agent prepares environments from natural-language descriptions. This suits research workloads like trajectory collection, reinforcement learning, and agent evaluation.
AX addresses a key problem: agents accumulate state, require strict isolation, call external APIs, and can waste money in loops without oversight. Google's four primitives aim to make agent orchestration as manageable as traditional cloud workloads. The declarative model lets teams suspend, resume, and delete tasks with simple commands.
Industry observers see AX as part of a broader push to operationalize AI agents in production. Scaling to billions of concurrent sessions per cluster could lower barriers for large-scale deployments. Still, adoption and ecosystem support will determine its success.
Google has not announced a general availability date, but the open-source positioning suggests a community-driven roadmap. If AX delivers on density and checkpointing, it could become a foundational layer for agent infrastructure. Developers can already experiment with the YAML workflow and generative workspace features.
