On October 8, Google Cloud announced an enterprise-focused Gemini agent at its Gemini at Work 2026 event, integrating Q&A, knowledge processing, content generation, and code writing and execution into a single entry point, while supporting connections to enterprise systems and tool calls to complete tasks.

The product is currently in private preview for enterprise customers and is available through the Gemini Enterprise app.

From goal-setting to work delivery

According to the product design Google released, users can assign goals to Gemini, and the agent plans the steps, selects tools, and executes them, delivering results to the documents, email, and development environments employees use daily. It also supports scheduled tasks and event-triggered workflows.

Google cited two office scenarios: when a manager requests a project progress presentation by email, Workspace can recognize the task and provide a one-click option to hand it to Gemini; when scheduling a team meeting, the agent can identify participants based on chat group members and previous activity emails, check the calendar, and then coordinate a time by email. Users can invoke the agent by mentioning Gemini in apps such as Gmail, Docs, Sheets, Slides, and Chat. The product also supports access via web, mobile, desktop, and channels such as Slack and Microsoft 365, and can be integrated into third-party apps through APIs.

Continuous execution in the cloud, with subtask delegation

Gemini runs in the cloud, preserving the same context and memory across devices. Google said that even if a user closes their laptop, work that takes hours or days can continue. For multi-step tasks, Gemini can dynamically create temporary specialized agents and coordinate them to work sequentially or in parallel. It can also take on ongoing team roles, such as project manager or financial analyst. These team agents can have independent identities, corporate email, and storage space, and colleagues can add them to work groups or mention them in documents. Google said team agents can only access content shared with them by users or team members.

Although the product name uses Gemini, the underlying model can be selected by task. It currently supports Google's Gemini series and Anthropic's Claude models, and Google plans to further integrate other commercial and open models.

Enterprises can set permissions, audit logs, and spending caps

Regarding agent access to enterprise systems, Google has set up independent identities and fine-grained permission mechanisms. Actions performed by agents enter audit logs, code and tasks run in isolated environments, and network traffic is managed by enterprise policies through Agent Gateway. On cost control, enterprises can set hard spending caps for projects in the cloud billing console. The system tracks model usage and runtime environment costs, and when the cap is reached, the project's agents are paused, with administrators able to choose to resume execution.

Google said nearly 80% of Google Cloud customers currently use its AI products, and nearly 90% of Fortune 100 companies use Gemini Enterprise. The new work agent will advance on top of this enterprise application platform, and Google plans to expand access to some Workspace Business and Enterprise plan customers.