AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Google Open-Sources AX a Kubernetes Style Orchestrator for Autonomous AI Agents

Collected Sep 30, 2026

Google has open-sourced AX, an orchestrator and declarative runtime for executing and scaling autonomous AI agent workloads. The project is hosted at agentexecutor.io and on GitHub as google/ax, and is licensed under Apache 2.0. The report is by Olimpiu Pop.

AX runs on top of Agent Substrate, an execution runtime engineered for dense actor multiplexing. Each agent session runs as an isolated actor sandbox with strict CPU and memory resource boundaries. When an agent goes idle, such as while waiting on an inference provider or a tool call, the platform checkpoints its execution state and suspends it. AX is designed to resume suspended actors in sub-second intervals with zero cold-start delay, multiplexing dozens of tasks onto shared host workers. AX treats agents as stateful actors rather than microservices or batch jobs.

The control plane exposes four declarative Kubernetes-style primitives under the ax.io/v1alpha1 API group. Task defines the execution lifecycle, sandbox resource constraints, and references to supporting infrastructure. Workspace handles pre-execution environment assembly, including mounting Git repositories, configuring Model Context Protocol (MCP) servers, installing skill bundles, or providing natural-language goals an initialisation agent executes to bootstrap toolchains and dependencies. Gateway manages outbound network security policies, restricting sandboxed agents to explicit allowlists of hostnames and ports while handling credential injection into outbound requests. Model is a unified control point for LLM provider parameters, runtime configurations, and secrets stored in Kubernetes.

The ax command-line tool is written in Go. Operators deploy the control plane to Kubernetes using ko and Redis into the ax-system namespace, interfacing through existing Kubernetes contexts via kubectx. Commands include ax apply to register manifests, ax watch to stream task phase and condition changes, ax ssh for interactive sandbox debugging, and ax suspend and ax resume for manual control of task state. The architecture is rooted in systems research from teams across Google and Google DeepMind.

Discussion on Hacker News was divided. According to the report, some infrastructure engineers praised the platform for addressing idle-agent cloud costs, while some developers criticised the marketing claim of ergonomic workflows because of operational overhead such as maintaining Kubernetes clusters, container registries, and custom CRDs via ko. Discussions cross-posted to Reddit emphasised that AX is a foundational execution runtime rather than a high-level application orchestrator like LangGraph or CrewAI, while practitioners highlighted gVisor-isolated sandbox blast-radius containment alongside early issues such as egress proxy dropped connections and rudimentary secrets management.

Read at InfoQ · AI, ML & Data Engineering

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Google has open-sourced AX, an orchestrator designed for managing autonomous AI agent workloads. AX operates on a runtime, Agent Substrate, treating agents as stateful actors. It provides resource-efficient task suspension and resumption to optimise performance and reduce latency in idle phases. It features a control plane with Kubernetes-style primitives for managing agent tasks and resources. By Olimpiu Pop