Google gives its Gemini agent an inbox, a calendar and a line in the company directory

At Gemini at Work 2026, Google introduced an agent that works in Workspace like a full-fledged colleague. It has its own account, email address, calendar and entry in the corporate directory. It runs tasks for days, including when the user's laptop has long been closed. Every action goes into the log under the agent's name, not under the name of the employee who launched it.

An agent with its own identity

Until now, most AI assistants in office suites have acted on behalf of the user. They read the user's email, edited their documents and sent messages from their address. In the audit log, this looked like the person's own actions, and separating their decisions from the machine's was almost impossible.

Google chose a different approach. The agent gets the same attributes as a new hire: a user account, a mailbox, a calendar, a place in the org chart. You can email it, invite it to a meeting, look it up in the address book. For an administrator, it is one more object to manage, like a user or a service account.

The second part of the announcement concerns uptime. The agent is not tied to an open browser tab or a powered-on device. It keeps carrying out assignments in the background and comes back with a result when it is ready.

Why Europe was waiting for this

The original TNW article stresses that tying the log to the agent is exactly what European regulators have been asking about all year. The logic is clear. If an autonomous system sent a customer an email or changed data, an auditor wants to see that the agent acted, on whose instructions, and with what permissions. When everything is recorded under a person, accountability blurs. When the agent is separate, there is a trail that can be reviewed and challenged.

This is our interpretation, not a statement from Google. But the direction is clear: an agent's identity is turning from a technical detail into a compliance argument.

What is known about the capabilities

• A separate Workspace account for each agent
• Its own email address
• A personal calendar
• An entry in the company's corporate directory
• Autonomous operation for several days, including while the user's laptop is closed
• The action log is tied to the agent, not to the person

What is left out

The available excerpt does not include pricing, regional launch dates, a list of supported plans or the access-rights model. It is unclear whether the agent will be able to correspond with external domains, who in a company gets to create one, and how shutting one down works. Whether the feature becomes a working tool or stays a demo depends on these details.

Practical takeaways for IT departments

Workspace administrators should think through rules for these accounts in advance. Each agent needs a designated owner, a limited set of permissions based on the principle of least privilege, and a separate policy for external correspondence. Agent logs are worth including in general monitoring and reviewing as regularly as the actions of privileged users.

The offboarding scenario matters just as much. If an agent was created by an employee who is leaving the company, its account needs to be transferred or closed. Otherwise the directory will keep a "colleague" with no owner. Companies with European divisions should show their legal and data-protection specialists exactly how the agent's actions are recorded, before the feature goes into broad use.