Declare an agentic job. AX runs it at scale.

About

Born from analysis, constructed for manufacturing.

AX was born at Google when agentic runtime techniques analysis met frontier compute. Over years
of constructing and working agentic execution engines, groups throughout Google acknowledged that agentic workloads
characterize a wholly new computing paradigm: stateful, bursty, long-running actors that compute intensely for a
minute after which look forward to mannequin responses, device responses, or human approval. Traditional orchestrators constructed for
stateless
microservices or predictable batch jobs turn into cost-prohibitive when conserving idle sandboxes working, but lack
native help for sub-second droop and resume.

Drawing on agentic runtime analysis from Google DeepMind alongside deep expertise in
large-scale isolation, resumption, and scheduling, AX is being constructed as an open, declarative management airplane
purpose-built for agent execution. It abstracts duties, workspaces, community insurance policies, and fashions into core
primitives so builders and researchers can run huge fleets of brokers with out reinventing the underlying
utilities. This undertaking closely depends on Agent
Substrate however gives agentic abstractions and generative runtime parts.