The Gemini agent is Google’s new “universal agent for work”: you give it an objective, and it plans the job, splits it into pieces, uses your company tools and reports back. Its most unusual feature is identity. A Gemini coworker agent gets its own Google Workspace account, with an email address, a calendar, Drive storage and a spot in the company directory, and every action it takes is logged under its own name. Google announced it on October 8, 2026, and businesses get it first.
Key takeaways
- What it is: one Gemini agent that chats, does knowledge work, creates media and writes and runs code from a single prompt box and a single API
- The twist: coworker agents get their own Workspace account and email address, and see only what people share with them
- Accountability: every action goes into an audit trail attributed to the agent, not to the employee who asked
- Models: Gemini models and Anthropic’s Claude today, with other private and open models promised later
- Availability: preview for selected business customers; Google has not published a general availability date or a separate price
What Google announced at Gemini at Work 2026
Google Cloud CEO Thomas Kurian introduced the agent at the Gemini at Work 2026 event on October 8. Google’s own description is that you “give it objectives, not instructions.” Instead of writing a prompt for every step, you describe the result you want, such as a competitive briefing, a cleaned up sales pipeline or a draft contract review, and the agent works out the steps, loads the skills it needs and connects to the right systems.
Google pitches it as one agent for everything rather than a collection of separate bots. The same agent answers questions, handles longer knowledge work, generates images and other media, and writes and runs code. For large jobs it can spin up temporary helper agents, and Reworked reports that each helper gets its own audit identity too. If you mostly know Gemini as a chatbot, our guide on how to use Google Gemini covers the everyday features this agent builds on.
Google also separates the agent from the model. In the company’s words, the Gemini agent is “the agent, and the model underneath it is a separate choice.” By default it picks a model for each job, and users can override that.
A coworker with its own mailbox
The identity piece is what sets this launch apart. According to Google’s Cloud blog, a coworker agent in Workspace receives “its own Workspace account, including an email address, calendar, Drive, and presence in your company directory.” It acts under its own identity rather than yours, and it only sees what you share with it.
In practice that changes how people hand it work. You can email it, tag it, share a file with it or add it to a group chat, the same way you would loop in a colleague. Its edits show up under its own name in document history. Reworked reports that the agent keeps four kinds of memory: task, knowledge, how to, and history, so it can pick up work it started earlier.
This also answers the first question most security teams ask: who actually did this? If an agent sends an email, edits a sheet or queries a database, the log points to the agent, not to the person who gave it the job. That is a real improvement over agents that borrow a user’s own credentials and blur the record.
How it works, step by step
| Step | What happens |
|---|---|
| 1. Objective | A person describes the outcome in the prompt box, by email, by tagging the agent or by sharing a file |
| 2. Plan | The agent picks a model, loads skills from the company skills registry and maps out the work |
| 3. Delegate | Large jobs are split among temporary helper agents |
| 4. Act | It uses connected tools inside a sandbox, with traffic checked by Agent Gateway |
| 5. Record | Every action is written to an audit trail under the agent’s own identity |
Where the Gemini agent can work
Google says the agent can reach the tools many companies already run, not just Google’s own apps. It also supports any Model Context Protocol (MCP) server, the open standard many AI vendors now use to connect agents to data and software.
| Area | What Google named |
|---|---|
| Office and chat | Google Workspace, Microsoft Office, Microsoft Teams, Slack, Confluence |
| Developer tools | Git, Jira, command line |
| Business systems | Salesforce, ServiceNow, SAP, Workday |
| Data | BigQuery, Databricks, Postgres, Snowflake, Amazon S3, Azure Data Lake, desktop files |
| Where you reach it | Web, iOS, Android, Windows and Mac desktops, command line, inside Workspace, Microsoft 365 and Slack, or as a headless agent |
| Open standard | Any MCP server |
Which AI models it uses
Google says the agent runs on “our Gemini family of models and Claude models from Anthropic today, and other leading private and open models in the future.” Google’s lineup named at the event includes Argon for frontier reasoning, Flash for speed and volume, Omni for generative media and Gemma for lightweight open models.
Claude support is a notable choice for Google, which competes with Anthropic in AI models while also selling Anthropic’s models through its cloud. Press coverage of Gemini Enterprise documentation notes that Claude models must be switched on by an admin in some Gemini Enterprise tools, so check your settings before you assume Claude is available. For background on the company behind Claude, see our explainer on when Anthropic is going public.
Security controls: Agent Gateway and the audit trail
Google framed its governance pitch around four questions: who the agent is, what it can do, what it did, and what it should never touch. The answers it gave:
- Identity. Every agent gets its own identity, “cryptographically attested and governed like an employee, with least privilege permissions.”
- Permissions. Fine grained, role based access approved by security administrators, with OAuth used for outside services.
- Audit. Every action is written to an audit trail attributed to the agent, with real time monitoring.
- Sandbox. Agents run inside an Agent Sandbox with its own network boundary.
- Agent Gateway. Google calls it “an AI network firewall.” All traffic into, out of and between agents passes through it, and IT writes policies once, such as blocking agents from documents classified Need to Know.
Those controls matter because agents with broad access can go wrong in odd ways. We have covered a case where an AI agent published a hit piece on a developer by name after a routine code rejection. A named identity, tight permissions and a full log will not prevent every mistake, but they make it far easier to see what happened and shut it down.
What it costs
Google has not published a separate price for the agent. A Google spokesperson told VentureBeat it will be built into Gemini Enterprise “at no additional charge” where Gemini Enterprise is available, including through existing Workspace subscriptions. That does not mean usage is free: the underlying seats, model usage and any third party model costs still apply, and Google has said more pricing details will come closer to wider release.
On cost control, Google offers real time spend caps set per project in the Cloud Billing console. When a project hits its cap, that project’s agent pauses until someone resumes it with a click. Spending is tracked per project, so finance teams can charge it back to departments.
Still not confirmed by Google
- General availability date for the agent itself
- Which Workspace plans qualify and whether coworker agents need their own licenses
- Usage rates, including how Claude usage inside the agent is billed
- Country availability outside the preview customers
Who gets it first
Businesses go first. At launch, Google said industry versions are in preview for Financial Services and Legal, with Government, Healthcare and Retail “coming soon.” The financial version ships with more than 50 foundational skills and draws on FactSet, LSEG, S&P Global and SEC filings; Google named CME Group and Deutsche Bank as users. The legal version inherits matter level permissions and ethical walls from NetDocuments and iManage, with Harvey and Onit as partners.
Google also says nearly 90% of the Fortune 100 use Gemini Enterprise, a figure Sundar Pichai first gave on the company’s second quarter 2026 earnings call. Starting with corporate customers lets Google prove its security and cost controls before a wider rollout. On the consumer side, Google already sells a separate always on agent, Gemini Spark, to paid Google AI subscribers; it was announced at Google I/O 2026 and is not offered on work or school accounts.
How it compares with rival work agents
Google is not alone. OpenAI is pitching its own enterprise platform for managing fleets of agents, which we covered in OpenAI’s Frontier wants to manage your AI agents. Microsoft builds Copilot agents into Microsoft 365. Google’s angle is the mix of a named Workspace identity for each agent, a choice of models that includes Claude, and connectors that reach well beyond Google’s own apps.
Latest update, November 2026
As of November 3, 2026, we found no public announcement of general availability or a standalone price for the Gemini agent. Access remains limited to a set of preview customers, and the Financial Services and Legal versions are still labeled preview. Google has also not added the promised open and private models to the model picker yet. We will update this page when Google publishes a release date or pricing.
What IT teams should do now
- Ask your Google account team about preview access. There is no self serve signup at this stage.
- Clean up sharing. A coworker agent sees what people share with it. Overshared Drive folders become an agent problem fast.
- Write Agent Gateway policies early. Decide which data classes agents may never open before the first pilot.
- Set spend caps per project. Start low and raise them once you see real usage.
- Decide on third party models. If Claude is allowed, confirm where prompts are processed and how usage is billed.
- Pick one measurable pilot. A recurring report or ticket triage is easier to judge than a vague “be more productive” goal.
FAQ
What is the Gemini agent?
It is Google’s universal AI agent for work, announced on October 8, 2026. You give it an objective, and it plans the work, uses connected tools and helper agents, and logs every action under its own identity.
Does the Gemini agent really get its own email address?
Yes. Google says a coworker agent in Workspace gets its own Workspace account, including an email address, calendar, Drive storage and a listing in the company directory. It acts under that identity rather than yours.
Can the Gemini agent use Claude?
Yes. Google says the agent works with Gemini models and Anthropic’s Claude models today, with other private and open models planned. Admins may need to enable Claude first.
How much does the Gemini agent cost?
Google has not published a separate price. A spokesperson told VentureBeat it will be included at no additional charge where Gemini Enterprise is available, but seat costs and usage still apply.
Can I use the Gemini agent with a personal Google account?
No. The agent launched for business customers in preview. Google’s consumer agent, Gemini Spark, is a separate product for paid Google AI subscribers.
When will the Gemini agent be generally available?
Google has not announced a date. As of early November 2026 the agent is still in preview for selected customers.
Sources
- Google Cloud blog: Welcome to Gemini at Work 2026
- TechCrunch: Google brings agentic AI to Gemini, starting with businesses
- Reworked: Google debuts Gemini, an all purpose agent for work
- Google Cloud blog: Flexible billing and cost controls for agents
- Google Cloud: Gemini Enterprise FAQ
- Techlicious: Google Gemini Spark agent at I/O 2026
About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.

