On October 8, Google Cloud announced the Gemini agent, a single AI agent for work that answers questions, handles tasks, creates content and writes code from one prompt box. The Gemini agent is Google’s answer to the wave of always-on agents that arrived this fall, and it comes with an unusual detail: it gets its own email address and its own Workspace account, like a new hire.
Here is what Google actually announced, how it differs from the chatbot you already know, and what is still unclear.
What the Gemini Agent Actually Is
According to TechCrunch’s coverage of the launch, Google Cloud CEO Thomas Kurian described an agent that takes “objectives, not just instructions.” You hand it an outcome, it plans the work, uses skills and tools, connects to company systems, and returns something finished.
Google’s own launch post lists the core design ideas:
- One agent for everything: chat, autonomous tasks and code generation all live in the same interface.
- Runs in the cloud: it keeps one set of memories and context across devices, and long jobs keep running after you close your laptop.
- Sub-agents: it can spin up temporary, job-specific agents and coordinate them on tasks that last hours or days.
- Model choice: it picks the best model for each job, currently Gemini and Anthropic’s Claude, with other models promised later.
An AI Coworker With Its Own Inbox
The most striking part is identity. Per TechCrunch, the agent has its own Workspace account, so it has an email address and its own context. It knows who is on which team, what time zones people work in, who approves what, and what is on their calendars. You can call it by tagging it, emailing it, sharing a file with it or adding it to a group chat.
Google also describes “coworker agents” with a persistent role, such as an analyst in a finance department. Reuters reports that these have their own email address and can only see the information they are given. When the agent acts, it writes an audit trail attributed to the agent rather than to a person.
That last point matters more than it sounds. When an AI sends an email or edits a spreadsheet under your name, nobody can tell who did what. A separate identity makes every action traceable, which is the kind of control companies ask for before they let software act on its own.
Where You Can Use the Gemini Agent
The agent works inside Gmail, Docs, Sheets, Calendar and other Workspace apps. It is also reachable through Microsoft 365, Slack, ServiceNow, the command line, iOS and Android, and Windows and Mac desktops. TechCrunch adds that it connects to systems such as Jira, Confluence, Git, BigQuery, Databricks, Postgres and Snowflake, and to any Model Context Protocol (MCP) server inside or outside a company’s network.
A “tasks inbox” lets you watch what it is doing, including its reasoning, the sub-agents it delegated to, the skills it loaded and its progress. Versions for financial services and legal work are in preview, and versions for government, healthcare and retail are listed as coming soon.
How It Fits the Agent Race
Google is not alone. Reuters notes that OpenAI launched its always-on “dots” agents in September, and Meta released Muse earlier that month. We covered Meta’s Muse launch when it happened, and our explainer on what AI agents are and how to get started covers the basics if the term is new to you.
Google’s pitch is distribution. Sundar Pichai said at the event that Gemini has over one billion monthly active users, and Google says nearly 90% of the Fortune 100 already use Gemini Enterprise. Early testers included On, Shopify and PayPal. Google plans to focus on businesses first and bring the agent to consumers later.
Including Claude as a selectable model is also a notable choice. It echoes a theme from our comparison of Claude and GPT-6 Astra: the best model changes every few months, so tools that let you switch are easier to live with.
What to Watch Before You Rely on It
Every claim above comes from Google’s announcement and press coverage of it. Independent testing of how reliably the agent finishes real work is not out yet, and a demo is not a deployment. Three questions are worth tracking:
- Reliability: does it finish multi-day jobs without drifting off task? Our piece on why agents forget things explains why persistent memory is the hard part.
- Cost: Google promises multi-model routing and real-time spend caps, but pricing details were not part of the coverage we reviewed.
- Consumer timing: Google has said business customers come first, with no date given for everyone else.
If you run a team, the practical takeaway is to look at how the agent handles permissions and audit trails before you look at its demos. The Gemini agent’s biggest idea is not that it can do more. It is that it can be managed like a colleague, with a name, an inbox and a record of what it did.
Sharing Is Caring