AI NewsModels & agentsAnnouncement

Google gave its new Gemini agent its own Workspace email and calendar

Google announced a Gemini agent for Workspace customers that gets its own company email at @agents.company.com, keeps its own calendar and memory, and signs its own audit trail when it acts.

AI News

Editorial2 min read

LinkedInX

Why it mattersAn agent that signs its own actions can be reviewed under the same quarterly access policy a company already runs for employees, so security teams audit the agent as a user and revoke it as a user.

An agent that signs into work email as itself has a user record a security team can audit the same way it audits a person.

Google Cloud announced the Gemini agent at the Gemini at Work 2026 event, in a post by Thomas Kurian, CEO of Google Cloud. Google says Gemini will roll out to Gemini Enterprise customers first, then to consumers, and that nearly 90% of the Fortune 100 already use Gemini Enterprise.

What it is and how it signs in

Google says the Gemini agent is one agent for work that answers questions, runs assigned tasks, generates images and media, and writes code, from one prompt box. Thomas Kurian writes that users give it "objectives, not just instructions". The agent plans the work, chooses tools, calls skills, and connects to company systems to deliver finished work back inside Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar.

Google says a Gemini "coworker agent" has its own identity, with an @agents.company.com email address, its own persistent storage, and access only to context a person on the team gives it. Every action the agent takes is logged under the agent's own identity in the audit trail. Users call the agent by tagging it, emailing it, sharing a file with it, or adding it to a group chat.

What it connects to

Google lists built-in connectors for Google Workspace, Microsoft 365, Microsoft Teams, Slack, Jira, Confluence, Salesforce, ServiceNow, Git, BigQuery, Databricks, Postgres, and Snowflake. The agent also calls any Model Context Protocol server (the open standard that lets an AI model talk to a tool) inside or outside the company network, and Google is opening an enterprise tools and skills registry so teams can publish internal tools and reusable skills to the rest of the company.

Model choice and controls

Gemini picks a model for each job and starts with Google's own models and Anthropic's Claude models, with Google saying open and other private models will follow. The post names On, Shopify, and PayPal as early testers, and says PayPal already routes ten million multi-model requests a week across different models. Google states Bradesco cut document review from one hour to five minutes, with risk inconsistencies down 60%, and that BNP Paribas is deploying Gemini into an internal assistant for more than 65,000 employees. These figures come from Google's post, so a buyer should ask the named customer directly before quoting them.

Google says new spending controls include smart routing, multi-model orchestration, and real-time spend caps.

An agent with its own identity is a change to how a company grants access. Giving a bot its own user account rather than a human service account means the tool calls it makes, the files it touches and the emails it sends can be filtered, approved and revoked under the same policy the company already applies to a new hire. The access review a security team runs each quarter now covers an agent that may read every project it was ever added to.

Source

Primary source: Welcome to Gemini at Work 2026: Introducing the Gemini agent by Thomas Kurian, Google Cloud. Also reported by Sarah Perez at TechCrunch and by The Verge.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX