Google Ships the "Universal" Gemini Agent for Work: A Cloud Coworker With Its Own Email Address — That Routes to Claude When It's Cheaper-Better

Google Ships the "Universal" Gemini Agent for Work: A Cloud Coworker With Its Own Email Address — That Routes to Claude When It's Cheaper-Better

New subject (announced Thursday, October 8, 2026, at Google Cloud's Gemini at Work 2026 event). Google Cloud CEO Thomas Kurian unveiled the Gemini agent — a single, persistent, cloud-run agent for enterprise work that "can work across apps and devices in the background" (The Verge). It is Google's enterprise answer to OpenAI's Dots agent push and Microsoft's Copilot/Foundry stack, and Bloomberg framed it flatly as "ramping up competition with OpenAI, Anthropic and Microsoft for corporate customers."

What actually shipped (private preview, enterprise customers; broad availability planned for Workspace Business/Enterprise plans):

• One agent, every surface — Gmail, Drive, Docs, Slides, Sheets, Chat, Calendar natively; also reachable from web, iOS, Android, Windows, Mac, command line, and third-party apps like Slack and Microsoft 365. Cloud-run, so "Gemini will maintain the same context across all connected devices."
• The "coworker agent" pattern — per The Verge, Gemini can "operate as a 'coworker agent' with a dedicated identity and its own '@agents.company.com' email¹." Quartz adds the coworker gets its own Workspace account, calendar, and company-directory presence, scoped only to information teammates choose to share — not inherited org-wide access.
• Four memory types — session, semantic, procedural, and episodic memory, so context survives across tasks and devices.
• Multi-model routing — including Anthropic's. The agent "selects from multiple AI models depending on the task. It currently supports Google's own Gemini models and Anthropic's Claude models²." A Smart Routing capability assigns each workload "to whichever model balances quality and cost most favorably," with admin-set spending thresholds that auto-suspend the agent when hit.
• Industry editions in preview for financial services and legal; government, healthcare, retail "coming."

Why it matters — three reads:

1. The agent platform war is now three-sided in earnest. OpenAI shipped Dots at DevDay (Sept 29), Meta has Muse, Microsoft has Foundry agents — Google's move puts a persistent, identity-bearing agent inside the productivity suite where enterprise work already lives, with its own governance layer (cost thresholds, scoped access) aimed at the buyer, not the user.
2. Google is routing enterprise workloads to its competitor's models. Shipping Claude as a first-class backend inside the Gemini agent — with routing that optimizes for quality-per-dollar — is a remarkable concession that Gemini alone doesn't win every workload, and it makes Anthropic revenue a component of Google's flagship enterprise agent (Meta and Microsoft Quietly Pull Internal Work Off Claude — Weeks Before Anthropic's IPO shows the same dependence cutting the other way at Meta and Microsoft).
3. The "coworker with an email address" is the new deployment unit. After Manus's Cue gave each agent "its own dedicated email address, phone number, and mobile wallet" (Manus Closes $500M+ After Beijing Forced Meta to Unwind Its Acquisition — The "Singapore Washing" Precedent Now Has a Happy Ending for Investors, and a Hong Kong IPO Path) and OpenAI's Dots run persistently, the industry has converged on persistent identity as the agent product pattern — the same pattern regulators are now scrutinizing in the rogue-agent wave.