One Agent, Its Own Inbox: Google's Gemini Agent Gets a Gmail Address and a Job

At its Gemini at Work 2026 event on October 8, Google Cloud announced something that looks mundane on the surface but marks a structural shift in how AI enters companies: the Gemini agent, a “single, universal agent for work” that — when deployed as a team member — receives its own Workspace account, complete with a Gmail address, calendar, Drive storage, and a presence in the company directory.

The image Google is selling is deliberate: the agent doesn’t just chat with you. It onboards. “Gemini onboards itself the way a new hire would, learning you, your tools, and your team before it starts,” the company wrote in a keynote post authored by Google Cloud CEO Thomas Kurian.

What the Gemini agent actually is

Strip away the keynote language and the architecture is straightforward but ambitious. One agent, one API, one prompt box. You “give it objectives, not instructions,” as Kurian put it — “You delegate an outcome and come back to finished work.”

Six design principles underpin it:

• Unified agent. It answers questions, works autonomously on assigned objectives, and writes and runs code from a single interface. You can assign work, schedule tasks, or have it respond to events.
• Omnipresent access. Web, iOS, Android, Windows, Mac, command line, Google Workspace, Microsoft 365, Slack — and it operates headless, with no dedicated UI, via API for embedding in third-party apps.
• Persistent execution. It runs in the cloud with a single set of memories, context, and one personalization graph across every device and channel. Work that takes hours or days keeps running after you close your laptop.
• Multi-agent orchestration. Gemini dynamically creates a roster of temporary, job-specific sub-agents — each with their own identity — for multi-step tasks that run in parallel or sequence for hours or days. “Coworker agents” go further: persistent team members with defined roles and their own @agents.company.com email addresses and storage.
• Deeply contextual. Four kinds of memory: session (the task in front of it, even across days), semantic (a knowledge base it builds as it reads and works), procedural (how a job gets done, including skills it writes for itself), and episodic (everything it has done before).
• Model choice flexibility. The agent and the model are decoupled. It runs each job on whichever model fits best — orchestrating across Google’s Gemini family and Anthropic’s Claude models today, with other private and open models to follow. “The best model for the task is not always the largest one,” Google noted — and because the leading model changes every few months, keeping that choice open means your context, skills, and data stay put when it does.

That last point is quietly the most subversive. Google — a company with every commercial incentive to lock enterprises into its own frontier models — is explicitly telling customers that its agent will route to a competitor’s models when the job calls for it. Analyst Holger Mueller of Constellation Research flagged exactly this: “What stands out is the open approach — which allows enterprises to use already trusted and purchased LLMs as part of the automation scope of Gemini agent.”

Inside Workspace: three modes of presence

The Gemini agent works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, carrying the same memory, skills, and controls everywhere else. Inline, @Gemini invokes it in a doc or thread.

As a personal assistant, it arrives already briefed — knowing your calendar, your team, your projects, and how your documents relate. Google’s example: ask it to set up a meeting with “the usual team of regional event leads next week” without supplying a single name or email address, and it infers who those people are from chat-space membership and your last event thread, checks their calendars, and starts a coordinating email — external participants included.

As proactive delegation, Workspace Intelligence recognizes delegatable tasks. If your manager emails asking for a project update as a slide deck, you get a one-click option to hand it to Gemini. The same reasoning surfaces the inbox message that matters most, not the one that arrived last, and explains why.

As a coworker, it’s at its strangest and most interesting. Describe the role you need and Gemini creates it — an events coordinator that drafts a launch readiness document and posts it back to the group, or an agent tagged in a document comment that suggests an edit and replies in the comment thread, “appearing under its own name in version history.” It acts under its own identity rather than yours, sees only what you share with it, and access follows the sharing and membership your team already uses.

Governance: agents with employee-grade identity

For enterprises, the governance section of the announcement may matter more than the demos. Google frames agent governance around four questions: Who is the agent? What is it allowed to do? What did it do? And what should it never touch?

Every agent gets a cryptographically attested identity, governed like an employee, with least-privilege permissions. That identity is stamped into logs and into any VM spun up to run code on its behalf. Fine-grained role-based permissions are approved by security administrators, and identities map through OAuth when the agent touches external systems. Every action lands in an audit trail attributed to the agent, not to a person — with real-time observability to catch anomalous behavior.

The fourth question gets its own product: Agent Gateway, an AI network firewall where all traffic — in, out, and between agents — passes through and is checked against organization-wide policy. Write “agents may not open documents classified Need to Know” once, and it applies to every agent in the company. Agents also execute inside an Agent Sandbox with its own network boundary.

On cost: multi-model orchestration, Smart Routing that triages workloads to the cheapest adequate model, and real-time spend caps set in the Cloud Billing Console — if a cap trips, that project’s agent pauses until you resume it with one click, and per-project tracking enables chargebacks to departments. Google noted per-token prices have dropped 98% since 2024, but enterprise volume has exploded.

The data layer and customer proof points

The agent is grounded in a Knowledge Catalog that maps business definitions once (“net margin,” “addressable market”) so all agents use them; Smart Storage enriches unstructured objects in place; and a “borderless Lakehouse” lets Gemini query Amazon S3, Azure Data Lake, Salesforce Data 360, SAP, ServiceNow, and Workday without copying data or paying variable egress fees.

The scale claims are the usual keynote fare, but a few stand out: nearly 500 Google Cloud customers each processed more than one trillion tokens in the last year; ~80% of all Google Cloud customers use its AI products; ~90% of the Fortune 100 use Gemini Enterprise. Bloomberg Media lifted SQL query accuracy 63% by grounding data agents in the catalog. Bradesco cut document review from an hour to five minutes. SOMPO built over 10,000 custom agents across 34,000 employees. Orange Spain deployed more than 1,000 custom agents with zero IT bottlenecks.

Early testers of the model-routing capability include sportswear brand On, Shopify (which blends frontier models for millions of merchants), and PayPal — which routes 10 million multi-model requests every week.

Availability

The Gemini agent is in private preview now, with wide availability “soon” for Workspace customers on select Business and Enterprise plans. Industry specializations for Financial Services and Legal are in preview, with Government, Healthcare, and Retail coming. Beneath it all sits Google’s AI Hypercomputer stack, with the latest TPU 8i system delivering 80% better price-performance than the prior generation — and, in a footnote that deserves its own story, NASA’s Jet Propulsion Laboratory is running Gemma directly on a satellite in orbit, a first for a vision-language model.

Analysis: the org chart gets a new row

The real bet here isn’t any single feature. It’s that the unit of enterprise AI is no longer the chatbot or even the copilot, but the persistent organizational actor — something with an identity, permissions, an inbox, a calendar, and auditable actions. Microsoft is building the same thesis from the 365 side; OpenAI’s DevDay agent push and Anthropic’s enterprise tooling approach it from the model side. Google’s differentiator is that it owns the productivity suite where the work actually lives — and that it’s willing to route to Claude to win the account.

The risk is equally structural: an agent with its own directory presence is only as good as its governance, and Google is effectively asking enterprises to trust that identity management, policy enforcement, and audit trails designed for humans can be stretched to non-human employees at scale. If it works, the question every CIO gets in 2027 stops being “should we use AI?” and becomes “how many agent accounts do we provision, and who approves them?”