Google's New Free Tool Runs Billions of AI Assistants at Once
This could make AI helpers cheap enough to run by the thousand
AX runs agentic tasks at scale — sandboxing each one, wiring up its workspace, and fencing its network, with billions of tasks per cluster. Every task runs as a lightweight actor, so there are no orchestrator limits. Idle agents waiting on model responses, tool calls, or human input are checkpointed and suspended, then brought back in under a second with zero cold-start delay. Dozens of tasks share worker resources, turning idle waiting time into spare compute. Its four primitives — Task, Workspace, Gateway, and Model — handle isolation, setup, network policies, and config declaratively. AX was born at Google.
- AX is free and open-source, and it's plumbing for AI agents — programs that take actions, not just answer questions.
- Google says one cluster can handle billions of tasks, and paused agents wake in under a second, so you only pay while they work.
- Every agent runs in its own sealed sandbox with an approved-websites-only network, which limits runaway costs and data leaks.
Why It Matters
Cheaper, safer AI agents could mean faster service, lower prices, and more automation in everyday apps.