Google's Open Agentic Orchestrator
Google's AX introduces an open-source orchestrator purpose-built for AI agents, aiming to handle their unique stateful and bursty nature at massive scale. It promises declarative management, sub-second resumption, and generative features, positioning agents as a distinct workload type. HN's interest is piqued by the technical innovation, but tempered by skepticism over Google's open-source commitment and the project's long-term viability.
The Lowdown
Google has unveiled AX, an open-source "Agentic Orchestrator" designed to manage and scale AI agent workloads. This platform aims to address the unique challenges posed by AI agents, which behave differently from traditional microservices or batch jobs, being stateful, bursty, and capable of significant resource consumption.
- Agent-Centric Design: AX provides four core primitives for declaring and managing agentic tasks, handling sandboxing, workspace wiring, network fencing, and scaling.
- Scalability and Efficiency: Built on Agent Substrate, AX can handle billions of concurrent agent sessions, offers sub-second resumption for idle agents, and employs dense multiplexing to maximize resource utilization by sharing worker resources.
- Generative Features: The platform integrates generative AI, allowing users to describe workspace environments in plain English, which AX then automatically prepares.
- Broad Use Cases: It supports various agent types, from interactive coding agents and Jupyter notebooks to headless browser testing and reinforcement learning experiments.
- Google's Pedigree: Born from research at Google, including insights from DeepMind, AX seeks to provide an open, declarative control plane specifically for agent execution, abstracting away complex infrastructure.
Ultimately, AX endeavors to simplify the deployment, evaluation, and scaling of AI agents for developers and researchers, offering an infrastructure solution tailored to the new computing paradigm these intelligent workloads represent.
The Gossip
Googling Open Source's Future
A significant thread of discussion centered on Google's role and reputation in open source. Commenters questioned whether AX truly represents a commitment from 'Google' or is merely a project by Google employees. Concerns were raised about Google's track record of deprecating products, leading to skepticism about the long-term support and viability of AX, despite its Apache 2.0 license.
Kubernetes Kismet or YAML Yawn?
The declarative nature of AX, particularly its use of YAML for defining tasks and workspaces, drew comparisons to Kubernetes. Some commenters quipped about the 'k8sification of AI' as an inevitable trend or even a form of 'salary justification.' Others expressed a general fatigue with YAML as a configuration language, despite understanding its pragmatic choice.
Substrate vs. Scion Showdown
One detailed comment compared AX (and its underlying Agent Substrate) with another Google project, Scion. The commenter felt Scion offered a more integrative approach, working well with existing tools, whereas AX/Agent Substrate was described as a 'greenfield independent effort' that requires opting into a 'pretty huge stack,' raising questions about its overall flexibility compared to alternatives.
Sales Scams and AI Spam
A passionate commenter launched into a tirade against frustrating experiences with Google Cloud Platform's sales team, describing them as pushing generic, LLM-generated decks filled with 'agentic-foo bullshit.' This comment highlighted a disconnect between high-level AI marketing and the practical needs of technically proficient users, expressing exasperation with irrelevant pitches that escalate to management.