Google unveils universal Gemini agent for long-running workplace tasks

At its Gemini at Work 2026 event on October 8, Google introduced a universal enterprise agent that can research information, create documents, write code, coordinate other agents and complete tasks lasting from hours to days, while maintaining memory, business context and identity across applications and devices.

The agent is designed to operate through existing workplace interfaces, including Gemini Enterprise, Google Workspace, Microsoft 365 and Slack, rather than requiring a dedicated app. Google said the goal is to turn enterprise AI from a collection of assistants into a unified platform that can take on entire assignments. In addition to Google's own Gemini models, the agent currently supports Anthropic's Claude models, with support for more proprietary and open-weight models planned.

Google also detailed a "digital coworker" configuration in which an agent can be provisioned with its own Workspace account, email, calendar, Drive storage and directory presence. Employees can assign work to it like a colleague, and because execution happens in the cloud, tasks can be started on one device and resumed on another. Each agent receives a cryptographically attested identity, with permissions managed through enterprise security systems and actions logged in audit trails. The memory architecture spans session, semantic, procedural and episodic memory to help agents build understanding of company processes and knowledge.

The company has not yet announced separate pricing, general availability or a full rollout schedule for the universal agent. The launch comes amid intensifying competition among AI companies to become the central AI interface for enterprise work, with rivals including Microsoft, OpenAI and Anthropic making similar moves in recent weeks.