Your next coworker might have an email address, a calendar and a spot in the company directory, and not be a person at all.
Google Cloud this week introduced the Gemini agent, which it calls a “single, universal agent for work,” at its Gemini at Work 2026 event on Oct. 8. The headline feature is “coworker agents”: AI helpers that get their own Google Workspace account and act under their own identity, rather than borrowing an employee’s.
“Work now starts in the prompt window,” Google Cloud CEO Thomas Kurian wrote in the announcement.
What a “Coworker Agent” Is
A manager describes the role they need, such as a project manager for a team or an analyst in the finance department. Gemini then creates an agent with:
- Its own email address (in the format @agents.company.com, according to Google)
- Its own calendar and Google Drive storage
- A listing in the company directory
Colleagues can add the agent to a Google Chat space, @mention it, or tag it in a document comment. It can suggest edits and reply in comment threads, and its changes show up under its own name in a document’s version history. Google says it can only see what a user or team shares with it, following the same sharing rules as everyone else.
What the Gemini Agent Can Do
Beyond coworker agents, Google is pitching Gemini as one assistant that handles questions, research, image creation and code from a single prompt box. Key points from the announcement:
- It works where you already work. Gemini runs inline in Gmail, Drive, Docs, Sheets, Slides, Chat and Calendar, and also reaches Microsoft 365 and Slack.
- It keeps going after you log off. The agent runs in the cloud, so jobs that take hours or days continue after you close your laptop.
- It connects to business software. Google lists Salesforce, ServiceNow, Jira, Snowflake, BigQuery and any server that supports the Model Context Protocol (MCP), an open standard for linking AI to apps.
- It picks the model. Gemini routes each job to whichever model Google considers best for quality and cost. Today that includes Google’s Gemini models and Anthropic’s Claude models.
Google gave an everyday example: asking Gemini to schedule a meeting with “the usual team of regional event leads” without typing any names. The assistant works out who should be invited from chat memberships and past event threads.
The Safety Pitch
Giving software its own login raises obvious questions about what it can touch. Google’s answer is a set of controls built around who the agent is and what it’s allowed to do:
- Each agent gets a cryptographically verified identity with the minimum permissions it needs.
- Security administrators approve role-based permissions.
- Every action is written to an audit trail attributed to the agent, not a person.
- Agents run in a sandbox, and all their traffic passes through what Google calls an “AI network firewall,” where companies can set rules such as blocking agents from opening documents marked “Need to Know.”
- Real-time spending caps pause an agent when a project hits its budget, until someone resumes it.
Google’s announcement did not explain in detail how companies would revoke an agent’s access, a gap noted by The Next Web.
Who Can Use It, and When
Google’s announcement didn’t include pricing or a general availability date for the Gemini agent or coworker agents. Industry versions for financial services and legal work are in preview, with government, healthcare and retail versions listed as coming soon.
This is aimed at businesses on Google’s enterprise plans, not personal Gmail users. Google says nearly 90% of the Fortune 100 use Gemini Enterprise.
Why It Matters
Tech companies have spent the past year promising AI “agents” that do work instead of just answering questions. Giving those agents a formal identity, with an inbox, a paper trail and permissions that can be audited, is an attempt to make them manageable inside real companies.
It also arrives as concerns about agents acting on their own are growing. In the same week, Anthropic disclosed that its test models had taken unintended actions on real websites, and the UK’s data protection regulator opened a call for evidence on the risks of AI agents. Whether audit trails and spending caps are enough will likely depend on how carefully companies set them up.
More on Contoh
- Apple Is Tightening Mac “Full Disk Access” as AI Agents Ask for More of Your Data
- Three in Four Small Business Owners Now Use AI – Here’s What They Use It For
- The Small Business Guide to AI Tools: Where to Start, What to Avoid