Google Cloud has introduced a new AI agent for the workplace, simply called the Gemini agent. The company announced it on October 8 at its Gemini at Work event, and it says the agent can take a goal, plan the steps, do the work in the background for hours or even days, and come back with a finished result. The most striking part is that it can also join a team as a “coworker” with its own email address.
What the agent does
Google describes the Gemini agent as a single, universal agent for work. Instead of asking separate tools to answer questions, write code or create documents, a person gives it an objective, and it chooses the steps, uses connected business systems and returns finished work in documents, email and developer environments. It works inside Gmail, Docs, Sheets, Slides, Drive, Chat and Calendar, and it can also be used through Microsoft 365 and Slack. It is available on the web, on iOS and Android, and on Windows and Mac desktops, and it keeps working after the employee leaves the computer.
For bigger jobs it can create temporary sub-agents, each with its own identity, and coordinate their work in parallel or in sequence.
Coworker agents
The feature that stands out is the coworker agent. A manager describes a role, such as events coordinator, and Gemini creates a persistent agent with its own Workspace account. That account includes an email address on a company-specific agents domain, a calendar, Drive storage and an entry in the company directory. The agent sees only the information that a person or team gives it, and every action it takes is logged under its own identity.
In practice, this means colleagues can email an agent, invite it to a meeting or assign it a task the way they would with a person.
Controls and safeguards
Google says the agent layer is built with enterprise controls. Each agent gets an identity and role-based permissions, and its actions are recorded in an audit trail. Tasks run in a protected environment called Agent Sandbox, while an Agent Gateway applies network rules between agents and outside systems. Companies can also set real-time spend caps that pause an agent’s work when a project reaches its budget, which matters because an agent that works for days uses computing resources the whole time.
Which models it uses
The agent is separate from the model underneath it. At launch it can route each task across Google’s Gemini models and Anthropic’s Claude models, and Google plans to add other private and open models later. Google also introduced specialized versions for industries. The financial services and legal versions are in preview, and government, healthcare and retail are listed as coming soon.
What is still missing
Availability and pricing are the open questions. Reports say the agent is in private preview with selected customers, and Google has not published a general availability date or a price. Google executives reportedly told analysts to expect wider availability around the end of October or early November, but that timing was not part of the public announcement. A Google representative said the company does not plan to charge a per-seat license for coworker agents in Workspace, and that pricing details will follow closer to the wider release. How agent activity will be metered instead is still unknown.
The wider race
The launch comes as the major AI companies compete to sell agents that work on their own. OpenAI introduced always-on agents it calls dots in September, and Meta released its personal agent Muse earlier the same month. Google also has a separate consumer agent, Gemini Spark, which requires a personal account with a paid Google AI plan and does not support work or school accounts, a naming overlap that can confuse buyers.
Google put some figures behind the pitch: it says nearly 500 Google Cloud customers each processed more than one trillion tokens in the last year, that nearly 80 percent of its cloud customers use its AI products, and that nearly 90 percent of the Fortune 100 use Gemini Enterprise.
Questions to ask before adopting
Companies considering the agent have practical questions to settle. Who owns a coworker agent’s mailbox and the data in it? What happens to incoming mail when an agent is paused or removed? How are the agent’s actions reviewed, and who is responsible for a mistake? And how will costs be forecast before there is any usage history to rely on?
What this means
Giving an agent an email address and a calendar is a small step technically, but a big one organizationally, because it puts software on the org chart. Whether it works will depend less on the demo and more on pricing, reliability over long tasks and how well companies can control what their agents do.
