Gemini Enterprise: Internal Agents and AI Search for Employees

GCP Study Hub
November 21, 2026

Gemini Enterprise is the app that makes agents and AI search available to a company's employees. It connects to internal data in sources such as Microsoft SharePoint, Gmail, and Google Drive, and it can surface both low-code agents built inside the web app and custom agents built elsewhere, such as ADK agents running on Agent Runtime. For the Professional Agentic Architect exam, the main architectural distinctions around Gemini Enterprise involve how it accesses company data, how teams create or register internal agents, and where its scope ends compared to MCP, A2A, and customer-facing platforms.

Connecting company data: data federation versus data ingestion

Gemini Enterprise can reach repositories such as Microsoft SharePoint Online in two ways, data ingestion and federated search, and the choice affects where content is stored, how permissions apply, and how good search results are.

  • Data ingestion copies documents from the source repository into the Gemini Enterprise index, in the same way that an Agent Search data store holds indexed copies of the content it serves. The indexed copy then exists outside the original system.
  • Federated search, or data federation, answers questions without copying content into Gemini Enterprise. Gemini Enterprise retrieves information directly from the source, so nothing is stored in its index.
  • With a SharePoint data store configured for federated search, Gemini Enterprise queries SharePoint on behalf of the signed-in user. Returned results follow each employee's SharePoint permissions, so users only see documents they can already open.
  • Federated search comes with two trade-offs. Unindexed data can yield lower search quality than ingested data, and the query text is sent to SharePoint, where Microsoft's terms and privacy policies apply.

The two modes trade off in opposite directions. Ingestion offers stronger search quality and keeps a copy of the content in the index, while federation leaves content in the source system and accepts some loss in search quality.

Building internal agents with Workflow Builder

Not every internal agent in Gemini Enterprise needs custom engineering. Workflow Builder, the low-code builder formerly called Agent Designer, runs directly inside the Gemini Enterprise web app for teams that do not have developers.

Users build and update agents in Workflow Builder using natural language prompts or a visual editor. These agents can connect to company sources such as Gmail and Google Drive, run on a recurring schedule, and be shared within the organization. Two distinctions place Workflow Builder among Google Cloud's agent options:

  • The team that owns a Workflow Builder agent changes it directly through prompts or the visual editor. Agents and automations built in code need a developer to write and maintain them.
  • Workflow Builder serves a company's own employees inside the web app. CX Agent Studio builds customer-facing conversational agents, which respond when a customer starts a conversation rather than running on a schedule.

Publishing Agent Runtime agents with agents-cli

Custom agents built outside Gemini Enterprise, such as ADK agents deployed to Agent Runtime, can be registered with Gemini Enterprise so employees can use them from the Gemini Enterprise web app. Alongside the console, teams can complete this registration from the command line using agents-cli.

In agents-cli, Gemini Enterprise registration belongs to the publish command group through the gemini subcommand:

agents-cli publish gemini --agent-runtime-id AGENT_RUNTIME_ID --gemini-app-id GEMINI_APP_ID

The --agent-runtime-id and --gemini-app-id flags identify the deployed agent and the target Gemini Enterprise app. If you omit the agent's ID, the command reads it from the deployment metadata file created during deployment.

It helps to know how the main agents-cli command groups divide the work. The deploy command pushes an agent to a runtime, and its --deployment-target flag accepts agent_runtime, cloud_run, and gke. The infra single-project command provisions infrastructure in a project. The extension commands add, list, remove, and update extensions to agents-cli itself. Registration with Gemini Enterprise sits in the publish group.

Architectural boundaries: Gemini Enterprise, MCP, and A2A

A recurring theme on the Professional Agentic Architect exam is separating employee-facing access from agent-to-tool integration and agent-to-agent collaboration. Gemini Enterprise is a place for a company's own employees to use agents. When agents themselves need to reach tools or work with other agents, that traffic runs over protocols built for the purpose.

  • MCP is the protocol for connecting agents to tools and data. Plain operations such as database lookups and API writes are natural MCP tools, and a single MCP server can serve many agents, which keeps credentials in one place and means a change to a database or API is made once.
  • A2A is the protocol for collaboration between independent agents, including agents run by other companies. An A2A agent describes itself by serving an Agent Card at /.well-known/agent-card.json on its domain. This discovery document follows the A2A specification and lists the agent's skills, capabilities, supported interfaces with their protocol bindings, and the security schemes clients must use, so any compliant client can read it.

Keeping these roles separate helps prevent architectural mismatches. A2A is meant for agents collaborating on tasks, so a simple lookup does not need its own agent and reasoning layer. MCP handles plain tool lookups and writes, and Gemini Enterprise gives a company's own employees a web app for internal agents and search.

Our Professional Agentic Architect practice exams test Gemini Enterprise alongside MCP tool exposure and A2A agent discovery, with detailed explanations that drill these distinctions.

Get tips and updates from GCP Study Hub

arrow