Google Cloud has launched a universal Gemini agent for the workplace, designed to handle multi-step tasks across apps, manage files independently, and operate on behalf of individuals or entire teams. Unveiled on Thursday, October 8, 2026, the tool integrates directly with Google Workspace, Microsoft 365, and Slack.
Alphabet Inc.’s Google announced a new Gemini agent during the Gemini at Work 2026 event, introducing a centralized system intended to execute complex, multi-day assignments across enterprise environments. Rather than simply reacting to individual prompts within a single application, the software functions as a persistent digital worker equipped with its own corporate identity, memory systems, and storage.
Google Cloud Integrates Workspace With New Software
Because it runs in the cloud, the agent maintains consistent context across all connected devices.
For example, when a manager receives an email requesting a project update slide deck, Workspace Intelligence recognizes the prompt as a delegatable duty and supplies a single-click option to transfer the assignment to Gemini.
Meeting coordination has received similar automation. The system determines who those people are from the membership of the chat space and the thread from the last event, checks participant calendars, and starts an email thread to coordinate a time that works, even when external participants are involved.
Enterprises Implement Coworker Agents with Gateway Security
Instead of merely offering individual support, enterprises can implement the software as a coworker configuration where agents function as a team member acting on behalf of multiple people, such as a project manager within a group, or representing a specific corporate role, like a finance department analyst. Each agent receives a cryptographically attested identity and permissions managed through enterprise security systems.
Utilizing Envoy and Kubernetes as its foundation, the Agent Gateway incorporates standard Kubernetes Gateway APIs to maintain openness and extensibility. It natively parses Model Context Protocol (MCP) tool calls and Agent-to-Agent (A2A) agent messages, enabling protocol-layer security policy instead of just network-layer filtering. Combining with Model Armor, the infrastructure establishes a centralized defense against widespread vulnerabilities like prompt injection, tool poisoning, and sensitive data leakage.
It chooses the best model for the job, has built-in cost controls, and most importantly, has the security, administration and governance required by your company.
Thomas Kurian, Chief Executive Officer, Google Cloud
Platform Manages Memory and Third-Party Connectivity
Under the hood, the system dynamically assigns workloads based on model suitability. This approach is meant to deliver optimal quality and lower costs. Temporary sub-agents tackle multi-step tasks, communicating with each other to coordinate workflows that include parallel and sequential steps running for hours or days.
To maintain organizational oversight, the platform operates with four types of memory: session memory for tasks in front of it even across days; semantic memory as a structured knowledge base built from reading documents, talking to people, and working with other agents; procedural memory for how a job gets done, including skills it writes for itself; and episodic memory for everything it has done before. Through a provided API, the software can be embedded within third-party applications to run as a headless agent devoid of any required user interface. Connectivity extends to external databases and platforms including Salesforce, ServiceNow, Jira, Git, BigQuery, Databricks, Postgres, and Snowflake.
Google Cloud Withholds Pricing and Rollout Details
Google Cloud has not announced a separate price for the universal agent, specified whether every capability will be included in existing Gemini Enterprise subscriptions, or provided a general-availability date and comprehensive rollout schedule.
