Google Cloud introduces a universal Gemini agent for cross-application work

The early-access Gemini agent is designed to carry out delegated tasks across Workspace, coding and analytics while preserving context and applying enterprise permissions to its actions.

Google Cloud introduced a universal Gemini agent for cross-application work on 8 October 2026 at its Gemini at Work event, describing one assistant that can handle knowledge tasks, write and run code, analyse data and continue assignments across business applications. The proposed experience combines a common prompt interface with tools, memory and organisational permissions.

In the company’s official announcement, Google Cloud CEO Thomas Kurian said the agent can operate inside Google Workspace and connect to other workplace systems. A Google Cloud spokesperson told VentureBeat that the new agent entered private preview with selected customers on 8 October. Google’s small-business announcement also describes early access for SMB customers, with broader availability promised but no confirmed date.

Workspace tasks, coding and analytics from one agent

Google says Gemini will work directly in Gmail, Drive, Docs, Sheets, Slides, Chat and Calendar, using shared context rather than treating each application as a separate assistant. The company also lists integrations with Microsoft 365, Slack, Jira, Salesforce and databases, with support for Model Context Protocol servers and reusable company-defined skills.

For a multi-step assignment, the agent is designed to gather information, produce documents and carry work into the application where the result is needed. Google also announced specialised data skills: engineers could request PySpark code or help with pipelines, while business users could ask questions that generate reports using BigQuery and the Knowledge Catalog. These capabilities were presented as part of the new offering, not as proof that every preview customer can already access each function.

Persistent work and agents with team identities

Google describes cloud-based execution that can continue after an employee closes a laptop, preserving context across devices and work sessions. The agent can delegate subtasks to temporary specialised agents, including tasks that run in parallel.

A separate coworker configuration is designed for teams. Google says such agents can be assigned their own Workspace accounts, email addresses, calendars and Drive storage, enabling colleagues to work with them through shared documents or Chat. Access to company information is intended to follow the permissions granted to the agent, not blanket access to everything its colleagues can see.

Enterprise controls and a limited preview

Google says the architecture assigns agents distinct identities, records their actions in audit logs and uses role-based permissions, isolated execution environments and an Agent Gateway to enforce network and data-access policies. It also describes selecting models from its Gemini family and Anthropic’s Claude according to the task, with broader model support planned.

The announcement sets out how those controls are intended to work, but does not provide independently verified results for extended cross-application assignments. Google’s early-access positioning does not establish general availability for the new agent or every described capability. The company has not announced a firm date for broader rollout, and existing Gemini Enterprise products have separate availability.