Google used its Gemini at Work 2026 event on October 8 to announce what it is calling the Gemini agent, a single universal agent for work that takes objectives rather than instructions and runs persistently in the cloud while a laptop stays closed.
The pitch is consolidation. Enterprise AI deployments have multiplied faster than most organizations can govern them through 2026 — a chat assistant in the productivity suite, a coding tool for developers, separate agents embedded in customer relationship management and IT service management platforms. The Gemini agent is Google's attempt to fold those use cases into one interface backed by one memory, one governance layer, and one bill.
What it actually does
The agent handles knowledge work, content generation, code writing, and multi-step task execution from a single prompt interface. It can run for hours or days on a single objective, spinning up temporary sub-agents for individual pieces of the work. It can also create persistent coworker agents that hold their own Workspace accounts, email addresses, calendars, and directory entries, which means organizations will provision, permission, and audit agents through the same identity processes they use for employees.
The integrations cover the expected enterprise surface: Google Workspace, Microsoft 365, Slack, Jira, Salesforce, ServiceNow, Snowflake, Databricks, and any Model Context Protocol server. The governance model includes cryptographically attested agent identity, OAuth-propagated permissions, action-level audit trails, and an Agent Sandbox with its own network boundary.
The detail that stands out is model routing. The agent selects between Google's Gemini models and Anthropic's Claude models depending on the task, with OpenAI and open-weights models to follow. Google says the top ten to fifteen percent of a typical workflow needs a frontier model and the rest does not, which it claims cuts costs by roughly a third. Cost controls include real-time spend caps per project, with per-user caps announced as coming.
The enterprise agent market this implies
Google Cloud's framing — one agent front door instead of a growing collection of assistants — is a direct response to the fragmentation that defined the first wave of enterprise AI deployment. It is also a competitive position against Microsoft, whose Copilot strategy has embedded AI across Microsoft 365 at every tier, and against the specialist vendors whose value proposition depends on sitting beside an existing workflow rather than becoming the workflow itself.
The coworker agent is the most consequential piece for enterprise application buyers. An agent with its own identity, email address, and storage is not a chat interface with memory; it is a durable software principal that security teams can track, audit, and deprovision like an employee. Microsoft is moving in the same direction with agent identities in Entra. The infrastructure for managing a mixed human-and-agent workforce is being built before anyone has worked out the organizational implications of running one.
The AI agent question in the back office has spent most of 2026 as a deployment question. The Gemini agent makes it a procurement question, which is the point at which model routing decisions start to get made by IT rather than engineering.
The product launched in private preview for a select group of Gemini Enterprise customers, with general availability expected in North America by the end of October. Pricing was not disclosed at launch.
Topics aitechnologyenterprisegoogleagentsworkplace
