Google's AX Open-Source Runtime Fixes What LangChain Breaks in Production

Most agent frameworks are great for demos. They fall apart in production. Agent Executor (AX) is Google's answer to that gap: an open-source, distributed runtime designed from the ground up for the operational realities of running AI agents at scale. It handles the infrastructure plumbing that no one wants to write but everyone eventually needs.

As we push agents to do more, a new operational problem has surfaced: long-running agent workflows are fragile and incredibly hard to manage reliably and efficiently in production. AX is Google's attempt to fix that at the infrastructure level, not the framework level.

The problem no framework was solving

AX is specifically designed for long-running workflows: AI processes that can remain active for minutes, hours, or even days and must be able to handle human input, network interruptions, or system errors in the meantime. That's a fundamentally different problem than what LangChain or AutoGen were built for.

One industry observer put it bluntly: "The event log, snapshotting, single writer model, and connection recovery in Agent Executor are exactly the things SRE teams have been duct taping for the last year," noting that existing frameworks are great for prototyping, but fall apart in production once agents run for hours or days.

What AX actually does

AX is a distributed agent runtime that coordinates agentic loops, manages executions with event logging, and communicates with both local and remote actors. It is designed for reliability, with native support for recovery and execution resumption, even in complex distributed setups.

The five core capabilities Google built into it:

• Durable execution: AX includes durable execution features for agents, harnesses, tools, skills, and sandboxes. It uses an event log and snapshotting to allow workflows to resume after outages or interruptions, including human-in-the-loop confirmations.