Google wants Gemini to become a digital coworker that can receive a goal, organize the steps and deliver a finished result. At its Gemini at Work 2026 event, Google Cloud CEO Thomas Kurian introduced the new Gemini agent, designed to work across business apps and internal systems. Its most concrete feature is the option for companies to create persistent agents with their own identity, email address and Google Workspace account.
The agent can answer questions, handle knowledge work, create content, and write and run code. Kurian describes it as a tool given “objectives, not instructions”: Gemini plans the work, uses skills and tools, connects to company systems and returns the result in the environments employees already use. Tasks can be assigned directly, scheduled or triggered by an event.
Google plans for it to work across Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, as well as Microsoft 365, Slack and third-party applications. The enterprise systems named include Salesforce, ServiceNow, Snowflake and Confluence. The agent maintains context across tools and devices and can keep working in the cloud after a user closes their computer; Google says some assignments can run for hours or days.
Agents can have their own identities and tools
Image: Google Cloud Companies can also create a “coworker agent,” a persistent agent with a defined role and dedicated Workspace account. It gets an email address, calendar and Drive storage, along with a place in the company directory. Colleagues can bring it into Chat or mention it in documents; Google says it acts under its own identity and sees only the content shared with it.
For complex assignments, Gemini can create temporary sub-agents and give each one a specific task to complete in parallel or in sequence. The distinction matters: the main agent coordinates the work, while sub-agents carry out parts of the assignment. This moves beyond chat to a system that can act across multiple tools while following company permissions.
Persistent memory and coordination across tools are already central to Google Cloud's continuous-memory agent. The new announcement shifts the focus from retaining context to applying it to business tasks, with access governed by the organization's policies.
Gemini can also choose Anthropic models
Gemini is not a single model. Google says the agent can route each assignment to the best-fit model, choosing among the Gemini family and Anthropic's Claude models. The company plans to add other proprietary and open-source models in the future. This selection layer, known as routing, lets the agent divide work among different models instead of sending every request to the same system.
Cost controls are part of the pitch: administrators can set a budget before work begins and apply company rules to agents. Kurian said an agent's identity is separate from an employee's, and that policies set for people also apply to agents. Spending remains a practical concern for businesses: a task may call different tools and models, while its budget must cover the work as a whole.
The announcement fits a broader shift from assistants that answer to agents that execute, a direction already visible in Google's strategy for enterprise agents. At the event, Sundar Pichai said nearly 80 percent of Google Cloud customers use AI products and nearly 500 enterprise customers each processed more than one trillion tokens on the platform over the last year. Those are figures Google reported about its own business. Google has not announced a price or a general availability date for the new agent.
No comments yet. Be the first!