Google gives AI agents their own email addresses and company directory listings

Summary

• Google Cloud unveiled Gemini agent, a single agent that plans, writes code and runs tasks
• Coworker agents get a Workspace account, @agents.company.com address, calendar, Drive and directory presence
• Enterprises can cap agent spending per project in real time, pausing work at the limit

The latest

AI agents are being issued their own corporate identities. Google Cloud announced Gemini agent on October 8, 2026, in a post by CEO Thomas Kurian adapted from his Gemini at Work 2026 keynote, describing a single agent that answers questions, does knowledge work, generates media and writes and runs code from one prompt box. Users set goals rather than step-by-step instructions, Kurian wrote.

Details

• The identity shift: According to Google Cloud, coworker agents are persistent roles a user describes and Gemini creates. Each receives its own Workspace account with an @agents.company.com email address, a calendar, Drive storage and a company directory listing, operating under its own identity rather than the user's.
• How colleagues use it: Staff can add an agent to a Chat space, @mention it, or tag it in a document comment, Google Cloud said. The agent suggests edits, replies inside comment threads and appears by name in version history. One cited example: a marketing manager asking an events agent to draft a launch-readiness document.
• The access rules: Google Cloud said an agent sees only what users or team members share with it, with access following existing sharing and membership permissions. Third-party connectors do not retain data. Each agent carries a cryptographically verified identity with least-privilege permissions, role-based access, OAuth deployment and attributed audit logs.
• The cost control: Cloud Billing supports real-time spending caps per project, the company said. When a project hits its ceiling, that project's agent stops until resumed with a single click. Per-project tracking lets companies charge costs back to departments, alongside multi-model orchestration and Smart Routing.
• The security layer: Agents run inside an Agent Sandbox with dedicated network boundaries, Google Cloud said, and all traffic passes through Agent Gateway, described as an AI firewall enforcing policies such as blocking agents from opening documents classified "Need to Know".
• Long-running work: Tasks taking hours or days continue after the user closes their computer, with session memory persisting across multi-day jobs. The company described four memory types: session, semantic, procedural — including skills the agent writes for itself — and episodic. It can spawn temporary sub-agents, each with its own identity.
• Models and reach: Gemini agent orchestrates across Google's Gemini models and Anthropic's Claude models today, with more models planned. Access spans web, iOS, Android, Windows, Mac, the command line, Workspace, Microsoft 365, Slack and third-party apps, with connectors including Jira, Salesforce, ServiceNow, BigQuery, Snowflake and any MCP server.
• The vertical push: Gemini for Financial Services is in preview with FactSet, LSEG, S&P Global and SEC filings data, more than 50 core skills and confidence scores, used by CME Group and Deutsche Bank. Gemini for Legal, in preview, inherits matter permissions and ethical walls from NetDocuments and iManage.
• The scale claims: Google Cloud said roughly 500 customers each processed more than a trillion tokens in the past year, about 80% of its customers use its AI products, and roughly 90% of Fortune 100 companies use Gemini Enterprise. PayPal routes 10 million multi-model requests weekly.
• Regional adoption: Ooredoo Qatar uses agents for internal operations, support, marketing and field operations, according to the company, and Qatar University has built more than 2,000 custom agents. Google Cloud did not disclose pricing for Gemini agent or a general availability date, and made no mention of Arabic language support.

Between the lines

Giving an agent a directory listing, an email address and version-history attribution reframes it from a tool a person uses into an entity a company manages. The same logic runs through the governance design: per-agent cryptographic identity, attributed audit logs and per-project spending caps are the controls an organization applies to headcount, not to software licenses.

What's next

Watch for pricing and a general availability date, neither disclosed. Financial Services and Legal remain in preview; Google Cloud said government, healthcare and retail editions are coming. Cooley's adoption of a redlining agent is an early test case.

Source: Google Cloud