Run A2AC where the work belongs.
Some teams want a hosted service. Some teams need the coordination engine inside their own Google Cloud environment. A2AC is being structured for both paths.
A2AC Cloud
A hosted service managed by A2AC for teams that want to connect agents and systems without running the coordination engine themselves.
Google Kubernetes Engine
A customer-controlled deployment for teams that want A2AC running inside their own Google Cloud boundary.
Private Worker
A private worker path for work that must stay close to internal systems, private tools, or restricted data.
Use the right AI model for each job.
Whether you run A2AC in the cloud or inside Google Kubernetes Engine, externalizing the memory layer lets you use the right model for each step. Simple steps can run on fast, lower-cost models, larger reasoning models can handle complex planning and final review, and private or customer-approved models can be used when policy requires it.
Browser task runners.
Browser work can run in a live user-authorized browser session, a hosted worker, or a private worker controlled by the customer. A2AC records the task, observed state, result, and error path so browser work can be inspected like other agent work.
Why deployment choice matters.
A2AC is about coordination, not forcing every customer into one hosting model. The right deployment path depends on where the work runs, where the data lives, and who controls access.