Google's Universal Gemini Agent Has Its Own Work Email

Google is turning its enterprise AI assistant into a single, persistent worker. At its Gemini at Work event on Thursday, the company announced a "universal" Gemini agent available through the Gemini Enterprise app, The Verge reports. Users chat with the agent and assign it work from one interface, but the agent is designed to keep running in the background across apps and devices rather than living inside a single chat session.

The agent works directly inside Workspace apps including Gmail, Drive, Docs, Sheets, Calendar and others. Users can also interact with it from a mobile device, desktop computer, the web, and third-party apps such as Slack or Microsoft 365. Google says that because it runs in the cloud, Gemini maintains the same context across all connected devices, so a task started on mobile can continue on desktop or web without losing state.

For execution, Google says Gemini will automatically perform tasks using an AI model that best fits the job, and it can work with job-specific sub-agents to complete tasks. That arrangement points to an orchestration layer in front of specialized models rather than a single monolithic model handling every request.

Google is also giving the enterprise agent a distinct directory identity. It can operate as a "coworker agent" with a dedicated identity and its own "@agents.company.com" email address, making it a named entity that can be managed alongside human employees instead of a temporary tool tied to a user's login.

The enterprise agent is currently limited to enterprise customers in private preview. That contrasts with the consumer-facing Gemini Spark agent, which Google launched in May. Gemini Spark lets subscribers perform tasks across Workspace and on the web, but through a separate interface within the Gemini app rather than a unified agent layer.

AI Mastery analysis

The coworker identity is the most consequential change here. Previous productivity AI features typically inherit a human user's permissions during an authenticated session. A Gemini agent with its own @agents.company.com address has to be governed as a persistent non-human principal. That shifts the design problem from prompt safety to service-account security: what can this agent read, modify, or trigger on a user's behalf, and how is that permission scoped across Google, Slack, and Microsoft 365?

As we wrote in AI Agent Security Fails at Shared Surfaces, Not Model Logic, shared cross-application surfaces are where agent security tends to fail. This implementation extends those surfaces to the corporate directory itself. Security teams will need explicit RBAC policies for AI principals that can ingest Slack channels or mutate Microsoft 365 data, with separate controls for the job-specific sub-agents the primary Gemini agent can delegate to. The worst failure mode is no longer a bad answer in a chat window; it is an over-privileged agent changing shared files or forwarding messages without a human in the loop.

Google's automatic model selection and sub-agent routing also confirm that enterprise automation is being built as orchestration infrastructure, not as a single scaled-up model. The engineering focus is on deterministic delegation, state continuity, and permission boundaries across APIs. Each job-specific sub-agent becomes a separate service with its own failure domain, and the primary agent must route, retry, and audit those calls across third-party APIs. In that world, the agent is managed like a background microservice with lifecycle controls, audit logs, and revocable access—not like a chatbot.

Sources

• Reported by: The Verge AI: Google is launching a one-stop Gemini agent for your work tasks