Google Cloud used its Gemini at Work 2026 event on October 8, 2026, to launch Gemini Agent, a unified AI agent for work that the company says can be given objectives rather than step-by-step instructions. The agent is available now in private preview to a select group of customers, with wider availability promised but no specific date — and no pricing details — disclosed.
The launch is Google's bid to consolidate enterprise AI into a single interface. Rather than separate assistants for chat, coding, and document work, Gemini Agent is designed to plan, execute, and complete multi-step tasks across applications, then return finished work.
An agent with its own identity
The most distinctive feature is what Google calls the "coworker agent." Enterprises can give Gemini Agent its own independent enterprise identity — a Google Workspace account with an email address, calendar, Drive storage, and a presence in the company directory. Team members can then assign it work by tagging it in chat, emailing it, or adding it to a group conversation, much as they would with a human colleague.
The agent acts under its own identity rather than the user's, and Google says it sees only what is explicitly shared with it. Every action it takes is written to an audit trail attributed to the agent, not to a person.
Built for long-running, cross-application work
Gemini Agent runs persistently in the cloud, maintaining memory and context across devices and channels. Google says tasks lasting hours or days continue even after a user closes their laptop.
For complex jobs, the agent can dynamically create and coordinate specialized sub-agents, each with its own temporary identity. It works inside Google Workspace — Gmail, Drive, Docs, Sheets, Slides, Chat, and Calendar — and can also be reached through Microsoft 365, Slack, ServiceNow, and third-party applications, with a headless mode for background use.
It connects to enterprise systems including Salesforce, ServiceNow, Snowflake, Jira, Confluence, Git, BigQuery, Databricks, and Postgres, and can work with any Model Context Protocol (MCP) server inside or outside a corporate network.
Google describes four types of memory supporting this: session memory for the current task, semantic memory as a structured knowledge base, procedural memory for how work gets done, and episodic memory for past work records.
Model choice — including a rival's
Gemini Agent supports dynamic multi-model orchestration, selecting between Google's Gemini models and Anthropic's Claude models based on task requirements. Google says support for additional private and open-source models is planned.
The company paired this with cost controls aimed at enterprise buyers: Smart Routing automatically selects models based on task complexity, and administrators can set real-time, per-project spend caps that automatically pause agents when thresholds are reached.
Google Cloud CEO Thomas Kurian framed governance and affordability as the two factors that determine whether enterprise agent programs succeed or stall.
On governance, Google detailed cryptographically attested agent identity, OAuth-propagated permissions, action-level audit trails, an Agent Sandbox for isolated execution, and an Agent Gateway described as an AI network firewall that enforces corporate policies across agent traffic.
What Google says enterprises are already doing
Google reported that nearly 90% of Fortune 100 companies use Gemini Enterprise and that nearly 500 Google Cloud customers have each processed more than one trillion tokens. Gemini has more than 1 billion monthly active users, the company said. These figures describe existing Gemini products, not the newly announced agent.
Early testers of Gemini Agent include sportswear brand On, Shopify, and PayPal. Named Gemini Enterprise customers include BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers.
Google also announced specialized capabilities for data analytics, data engineering, and machine learning, along with preview versions for financial services and legal industries.
A crowded field
The launch arrives during a stretch of competing announcements. Microsoft introduced Copilot Autopilot on September 25, 2026; OpenAI launched its always-on Dots agents on September 29; and Anthropic announced native Claude integration with Google Docs, Sheets, and Slides on October 6 — two days before Google's event.
Keith Kirkpatrick, an analyst with Futurum Group, said Google is "arguably leading the conversation around identity, audit trails, and who, or what, is ultimately responsible for workflows" in the enterprise agent space.
What is confirmed, and what is not
Confirmed: Gemini Agent exists, is in private preview, will be included at no additional charge for customers where Gemini Enterprise is available, and supports coworker agents with independent Workspace identities, cross-application execution, persistent memory, multi-model orchestration, and governance controls.
Unknown: Pricing for the agent itself, general availability timing, how usage-based charges for long-running tasks and sub-agents will be metered, and whether persistent coworker accounts will require separate licensing. Google has also not published independent benchmarks showing how reliably the agent completes lengthy cross-application assignments, nor has it detailed data-handling arrangements when third-party models such as Claude are used.
Awaited: Pricing details and availability dates, which Google said it will share closer to wider release, along with delivery of the announced industry specializations beyond financial services and legal.