Google Cloud launches Gemini Agent, an AI tool that executes tasks in the cloud and maintains context across devices

On Thursday, 8th, Google Cloud announced the Gemini Agent, an AI solution designed to integrate research, content creation, programming, and corporate task execution into a single interface. The system can leverage both Google’s Gemini models and Anthropic’s Claude to perform operations.

The announcement was made by Thomas Kurian, CEO of Google Cloud, at the Gemini at Work 2026 event. According to the company, the agent plans task steps, selects appropriate tools, and accesses corporate systems to deliver results directly within the applications employees use.

“It plans the work, uses skills and tools, connects to your systems, and delivers a finished product—within the documents, inbox, and development environments where you already work,” Kurian said in a statement.

Functionality and models

Google distinguishes the agent from the models that power its operations. The Gemini Agent coordinates activities and preserves context, while the model responsible for each task may vary based on quality and cost requirements.

Currently, the platform supports both Gemini and Claude families. The company plans to incorporate additional open and private models in the future, maintaining the agent’s associated context, data, and capabilities.

The choice of model depends on processing costs. In many AI applications, billing is based on tokens, the units models use to process input and output content. Google states that the agent includes cost controls, but the announcement does not provide pricing details.

Continuity of cloud-based tasks

The Gemini Agent executes tasks in the cloud and maintains memory and context across devices and channels through which it is accessed. According to the company, tasks that take hours or days can continue even after the user closes their computer.

The announcement promises access via web, Android and iPhone mobile devices, Windows and Mac computers, and channels such as Google Workspace, Microsoft 365, Slack, and command line. The agent can also operate integrated with other applications without a dedicated interface.

Integration with corporate tools

Within Workspace, one example presented is market trend research, building a financial model in Sheets, and creating a presentation with the results, without the employee needing to re-explain the project at each stage.

According to Google, the system can also suggest tasks for delegation and prioritize inbox messages, providing explanations of their relevance.

For complex tasks, the Gemini Agent can create temporary subagents responsible for steps performed in parallel or sequentially. Another option is to configure a persistent agent as a team member, with a defined role, email address, and storage, as well as access to shared context. Employees could include it in chat spaces or mention it using @.

Governance, security, and auditing

According to Google, each agent receives an identity and permissions approved by the organization’s administrators. Recorded actions are linked to the agent in an audit trail, which can be monitored using the company’s monitoring tools.

Execution occurs in an isolated environment called Agent Sandbox, with network restrictions. The company states that access policies can be applied to the entire set of agents.

Results from Brazilian clients

In the announcement, Google also showcased applications of Gemini Enterprise in Brazilian companies. The results refer to the corporate platform, not necessarily to the new Gemini Agent.

At Bradesco, document analysis time for tax, accounting, and contractual activities decreased from one hour to five minutes. Google also reports a 60% reduction in risk inconsistencies.

At Globo, the PlanejaAI tool, developed on top of Gemini Enterprise, reportedly saved 127 hours in a quarter for software planning. The indicators were released by Google and were not accompanied by a measurement methodology.

The material does not detail the availability schedule for each feature or the commercial conditions of the Gemini Agent.

Frequently asked questions

• What is the Gemini Agent?

  It is an AI agent announced by Google Cloud to answer questions, create content, write and execute code, and perform tasks across corporate systems.

• Does it only use Google models?

  No. According to the announcement, it can select both Gemini and Claude models. The company plans to incorporate other open and private models in the future.

• Do tasks continue when the computer is closed?

  Yes. Execution occurs in the cloud, allowing long-running tasks to continue even after the user closes their computer.

• How does the company control the agent’s access?

  Google states that agents have their own identity, permissions approved by administrators, audit records, and execution in an isolated environment.

• How much does the Gemini Agent cost?

  The launch announcement does not provide pricing information or detail the availability schedule for each feature.