Google is unifying work agents in Gemini, for now only in private preview

Google introduced the Gemini agent as a single interface for answering questions, creating content, writing and running code and autonomously completing work objectives. It runs in the cloud, keeps context across devices and channels and is designed to operate through Gemini Enterprise, Google Workspace, Microsoft 365, Slack, the command line and third party applications.

One agent is meant to connect inboxes, documents, code and business systems

A user provides an objective and Gemini plans the steps, selects tools and returns finished work inside a document, inbox or development environment. It can schedule tasks, respond to events and continue jobs for hours or days after the laptop is closed.

For more complex work, it can create temporary helper agents. Google also describes a coworker agent with its own corporate identity, email, storage and limited access to supplied context. The underlying model is selected for each job, including models from the Gemini family and Claude models from Anthropic.

Google is turning the chat box into an enterprise control surface

The strategic move is the combination of context, permissions, tools and memory. A company would not need to assemble separate agents for email, data analysis and code. Google wants those capabilities to share one identity and one management layer across channels.

Its distribution advantage is substantial. Google says nearly 80% of Google Cloud customers use its AI products and nearly 90% of the Fortune 100 use Gemini Enterprise. It also says almost 500 cloud customers each processed more than 1 trillion tokens in the past year. These are vendor figures, but they show the target: existing contracts, directories and data systems rather than a standalone consumer app.

A universal interface also widens the reach of a single mistake

The product is currently available only to selected enterprise customers in private preview. The announcement therefore describes a direction and architecture, not a generally available service that any European team can switch on today. Independent operating data on long task success, pricing and human interventions is also missing.

Putting email, documents, code and tools behind one agent identity raises the stakes for permissions, audit and recovery. The apparent simplicity of one prompt may create a complicated access graph for administrators. A coworker agent with its own account needs boundaries as legible as an employee’s, or the company merely shortens the path from a bad objective to a real action.

Production audits will show whether one agent can carry a full shift

The decisive evidence will come from pilots beyond curated demonstrations. Buyers need completion rates without correction, counts of incorrect tool calls, approval paths for sensitive steps and a complete audit of which model or helper agent performed each action.

Cost control will matter just as much for jobs running several days and switching dynamically between models. If Google provides clear permissions, spending limits and safe task recovery, a unified agent could replace part of today’s fragmented internal automation. Without them, the single window mostly hides complexity from users and hands it to administrators.

Lilith's verdict

Gemini is getting a corporate email, calendar, memory and a shift that can last several days. Before the company gives it a badge, it should know exactly who cleans up a wrongly sent message or an executed script.