Scale AI and Google Cloud Publish Enterprise Agent Deployment Blueprint

Scale AI and Google Cloud have published a joint reference architecture for running Scale GenAI Portfolio agents on Google Cloud and surfacing them in Gemini Enterprise. The release aims to give enterprise teams a documented path from agent development to employee use, including the infrastructure, permissions and controls required along the way.

The companies announced the architecture at the Google Cloud Doha Summit on September 22. Scale frames it as an answer to a familiar deployment problem: getting an AI pilot working is only part of the task; operating it in production also requires decisions about enterprise data, access controls, infrastructure and ongoing evaluation.

A design for moving beyond the pilot

The division of labor is explicit. Google Cloud supplies the infrastructure layer, including access to Google and third-party models, identity and governance services, Google Kubernetes Engine, GPUs and TPUs. Scale supplies its GenAI Portfolio for agent development, orchestration and deployment, plus evaluation, tracing and human review.

Under the proposed pattern, a team builds and evaluates an agent in Scale GenAI Portfolio, then deploys it inside the customer’s own Google Cloud project using that customer’s virtual private cloud and encryption keys. It can then offer the same agent through a dedicated business application and through Gemini Enterprise, where employees can find it subject to the organization’s access permissions.

Controls travel with the deployment

The architecture uses A2A and MCP, two protocols for connecting agents to other systems, alongside an Agent Registry for discovery. Scale says those interfaces let teams make an existing agent available in another interface without rebuilding its core logic, while also making it easier to connect agents with enterprise data and with one another.

The shared governance layer includes

• Identity, data-loss-prevention, audit and spending controls applied through Google Cloud’s governance framework.
• Prompt-safety and network-protection controls for the deployed agent environment.
• Operational visibility and records intended to support audit and review.

That common layer does not remove the need to assess each agent’s particular job. The architecture is meant to provide a common foundation across applications, while teams still address additional requirements for an individual use case.

The next decision remains with the customer

Scale’s recommended starting point is a single workflow with a clear owner and a measurable outcome. Before launch, teams are advised to map data sources, access needs, evaluation criteria and deployment requirements, then define acceptable performance and how they will monitor it afterward.

Scale also says it can provide domain expertise, Arabic-language support and delivery teams that remain involved after deployment. The reference architecture offers a starting design; the practical next move is choosing the first workflow and proving that it meets the organization’s own operating standard.