Google introduced a new Gemini Agent on October 8 at its annual Gemini at Work 2026 event, pitching it as a single AI assistant that can take on employees’ day-to-day work from one place. The agent is designed to handle tasks spanning email, documents, meetings, reports, presentations and coding, working across Google’s services rather than answering one question at a time.
According to Google, the idea is that a worker can hand the agent a complete responsibility instead of a single prompt. The company framed the launch as a shift from AI assistants that respond to requests toward agents that carry out multistep work on a user’s behalf.
Working inside Workspace
The central feature is that Gemini can now act directly inside Gmail, Google Drive, Docs, Sheets, Slides, Calendar and Chat. In Google’s example, an employee who needs to set up a team meeting next week could have the agent identify the relevant people from past conversations, check their calendars and propose a suitable time. Gathering information on a project, drafting a report and turning that report into a presentation can likewise be treated as one continuous task.
Google has also given the agent persistent memory. The company says it can retain context from earlier conversations, projects, preferences and previously completed tasks. For work that runs over hours or days, the agent can keep operating in the cloud after the user closes a laptop.
Agents as team members
Another element of the announcement is the ability to treat an AI agent as a formal member of a team. Companies can create a dedicated agent for a specific role, with its own email address, calendar and files. Employees can add it to team conversations and assign it work, much as they would a colleague. Google says each agent’s access is limited to the data and services it has been explicitly permitted to use.
The system is not tied to a single AI model. Google says Gemini can select among different models depending on the nature and cost of a task, including Google’s own models and Anthropic’s Claude models. A cheaper model can be used for simple work and a more capable one for complex tasks.
Controls for enterprises
Google emphasized security and oversight features aimed at business customers. Each agent’s access can be restricted, its activity logged, and its reach into specific files or systems blocked. Companies can also set spending limits on AI usage to keep costs under control.
Google said its AI products are now used by roughly 80 percent of Google Cloud customers, and that about 90 percent of the world’s 100 largest companies use Gemini Enterprise. Those figures come from the company’s own announcement and have not been independently verified.
With the launch, Google is trying to turn its existing office software into an environment where an employee delegates work to an AI agent rather than opening a separate app for each task. Its success will depend on how reliable the agents prove in real workplaces and how much access to sensitive information companies are willing to grant them.
