Summary
Google Cloud has announced the Gemini agent, a cloud-running system designed to handle workplace tasks across connected tools and services. Google says it combines persistent context, multi-agent workflows, model choice and enterprise security controls.
Google Cloud announced the Gemini agent on October 9, 2026, describing it as a single system for workplace tasks ranging from answering questions and creating media to writing and running code. Rather than giving it step-by-step instructions, users can assign an objective; Google says the agent can plan the work, use connected tools and return a completed result in the applications people already use.
The announcement presents Gemini as an enterprise agent platform, not just a chat interface. Google says it can operate across web, mobile and desktop, as well as through Google Workspace, Microsoft 365, Slack and command-line channels. The company did not give a general-availability date for the core agent in this announcement.
How Gemini is designed to work
Google describes three building blocks for connecting the agent to an organisation’s work: tools, skills and context. Tools connect it to services and data sources such as Microsoft Office, Salesforce, Jira, databases and Model Context Protocol (MCP) servers. Skills are reusable instructions or workflows for particular tasks. Context includes information drawn from interactions, documents and organisational knowledge.
The system is designed to keep working in the cloud when a user leaves a device, retaining context across sessions and channels. Google says it maintains session, semantic, procedural and episodic memories. In practice, those categories are intended to cover the current task, organised knowledge, ways of completing work and records of previous activity.
For complex objectives, Gemini can create temporary, task-specific sub-agents and coordinate their work in parallel or in sequence. Google also describes persistent “coworker agents” that can take on a defined team role. In Workspace, these agents can have their own account, email address, calendar and Drive; Google says they act under their own identity and see only what a team shares with them.
Gemini is also designed to work inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. Google’s examples include coordinating a meeting using information from a team chat and calendars, then carrying a project across applications to produce a financial model and presentation.
Models, controls and rollout
Google says the agent can route work among its Gemini models and Anthropic’s Claude models, with support for other private and open models planned. The company describes this as a way to match the model to a task and manage costs. Its announced controls include Smart Routing, project-level real-time spending caps, and an Agent Sandbox for executing tasks. Agent Gateway is described as a network control that applies organisational policies to traffic between agents and other systems. Google also says agent actions are recorded in audit trails and associated with the agent’s identity.
The announcement includes domain-specific versions for financial services and legal work, which Google says are in preview. Specializations for government, healthcare and retail are described as coming soon. These are distinct from the core agent announcement: the preview status applies to the named industry offerings.