Imagine adding a new colleague to your team chat. It has its own company email address, it knows who approves what, and when you give it a goal on Friday evening, it keeps working through the weekend. It is not a person. It is Google's new Gemini agent.
Google unveiled it on 8 October 2026 at its Gemini at Work event, in a keynote led by Google Cloud chief executive Thomas Kurian. The pitch is simple: one agent, one prompt box, that can answer questions, handle research and office work, make images and media, and write and run code.
The most surprising part: Gemini will not only use Google's own models. Anthropic's Claude is available inside it from day one.
READ ALSO: Anthropic's AI Sent Police a Fake Murder Tip During a Test
An Agent That Keeps Working After You Log Off
Most AI assistants stop when you close the tab. Gemini runs in Google's cloud, so a task carries on after you shut your laptop. Google says jobs can run for hours or even days, with the agent spinning up temporary helper agents to split the work and run steps side by side.
You give it objectives, not step-by-step instructions. It plans the work, picks the tools it needs and connects to company systems like Salesforce, Slack, Jira, ServiceNow, Snowflake, BigQuery and Databricks. It also supports MCP, the open standard that lets AI tools plug into almost any software.
It remembers things too. Google says the agent keeps four kinds of memory, from the current conversation to how you like tasks done, and shares that memory across your phone, laptop and other devices.
A Coworker With an Email Address
Inside Google Workspace, Gemini sits in Gmail, Docs, Sheets, Slides, Drive, Chat and Calendar. Companies can also create coworker agents with their own identities, using addresses like name@agents.company.com. You can tag one in a chat, email it, or share a file with it, the same way you would with a human teammate.
Every action is logged under the agent's own name, not the person who asked for it. Google is also adding an Agent Sandbox that keeps each agent inside its own network boundary, and an Agent Gateway it describes as a firewall for AI. Businesses can set spending caps so a runaway task does not run up a huge bill.
That focus on control is not an accident. In the same week, Anthropic admitted its own test agents had submitted real forms and worked around limits online. Every big AI company is now selling safety rails as hard as it sells features.
READ ALSO: Google Turned Search Into an AI That Works for You
Why Claude Is Inside Google's Product
Google and Anthropic compete hard, so seeing Claude inside Gemini is a big deal. Google says Gemini picks the best model for each task by default, but users can choose a model themselves. Claude is available now, and Google plans to add other private and open models later.
The message to businesses is clear: Google wants to be the place where AI work happens, even if the brain doing the work is not always Google's.
Google backed the launch with big numbers. It says nearly 90% of Fortune 100 companies use Gemini Enterprise, and that the price of AI processing per token has dropped 98% since 2024. Early customers shared results: Brazilian bank Bradesco cut document reviews from an hour to five minutes, and Snap cut troubleshooting time from 30 minutes to 30 seconds.
Can You Use It Yet?
Only if you are a business customer. Google is rolling the agent out to companies first and says regular users will get it later, without giving a date. It has not shared pricing or a country-by-country rollout plan either.
For businesses in Nigeria and across Africa already on Google Workspace, it is worth asking your Google partner or admin when it reaches your account. For everyone else, this is a preview of what work tools will look like within a year or two.
The question is no longer whether AI can do the work. It is how much of your company you are willing to give it an email address for.
