Google announced on October 8, 2026 that Gemini is becoming an agent rather than an assistant: one interface that takes an objective, plans the work, calls tools and acts inside your business systems - while carrying its own corporate identity. TechCrunch has the full launch details. Businesses get it first; consumers come later, once Google says it has solved “the harder problems around security, scale, and performance.”
Key facts
| Fact | Detail |
|---|---|
| Question answered | Google Brings Agentic AI to Gemini: An Agent With Its Own Work Email |
| Short answer | At a Google Cloud event on October 8, 2026, Google turned Gemini into an agent that plans work, picks its own model (Anthropic’s Claude included), connects to MCP servers and business systems, and gets its own… |
| Reading time | ~4 min |
| Last updated | 2026-10-09 |
What Google announced
- Objectives, not instructions. Google Cloud CEO Thomas Kurian’s framing: the agent plans the work, uses custom skills and tools, and connects to internal systems to accomplish a goal.
- A model picker. It defaults to whichever model fits the task, but users can take over - including third-party models, starting with Anthropic’s Claude. Open source and private models are promised later.
- Attachments and connectors. Requests can carry files, folders or project bundles; connectors include Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres and Snowflake.
- MCP everywhere. It works securely with any Model Context Protocol server inside or outside the company network - the same protocol many desktop tools now expose.
- A tasks inbox. A single UI that shows the agent’s thinking, its delegation to subagents, skill loading, code and progress.
- Spend controls. Multi-model orchestration, smart routing and real-time spend caps are aimed at keeping enterprise AI costs bounded.
- Scale claims. Sundar Pichai said Gemini has more than 1 billion monthly active users and that nearly 90% of Fortune 100 businesses use Gemini Enterprise at work. Early testers named in the announcement include On, Shopify and PayPal.
The detail that matters most: the agent has its own identity
Notably, the AI will have its own Workspace account, as if it’s just another co-worker. That means its own email address and its own context: it knows who is on which team, their time zones, who needs to approve items and what is on people’s calendars. You can call the agent by tagging it, emailing it, sharing with it or adding it to a group chat - and as it acts, it writes an audit trail attributed to the agent instead of to a person.
That is the piece enterprises were missing. Automation whose actions are logged under a human’s name is an audit problem waiting to happen. An agent identity separates “a human approved this” from “the agent did this”, which is exactly the boundary governance teams ask about before letting automation near production systems.
What it means if you run agents on Windows
Three practical consequences:
- Desktop is now a first-class surface. Windows and Mac desktops are on the access list alongside iOS, Android, the CLI, Google Workspace, Microsoft 365, ServiceNow and Slack. “Agents only live in someone else’s browser” stops being a safe assumption.
- MCP is becoming the integration contract. If your internal tools speak MCP, a hosted agent and a local desktop agent can both reach them. The difference is where execution happens and who holds the credentials.
- Governance moves to the agent, not the model. Model choice is being commoditized - Claude running inside Gemini is the proof. What differentiates platforms now is identity, permissions, approval steps and logs.
MOOGH sits on the other side of that line: it runs on your own Windows 11 machine, proposes a plan, waits for your approval and then executes locally - file operations, commands and repetitive automation under your control rather than under a vendor’s. To try that local-first model, download the client.
What to watch
- Whether the model picker really ships open source and private models, or stays a roadmap promise.
- Whether the coming consumer rollout inherits the same security posture Google is using to justify the business-first order.
- Whether an “audit trail attributed to the agent” becomes a portable standard or stays inside Google’s stack.
FAQ
- Is this the same as an agent running on my PC? No. Google’s agent is hosted and reaches your systems through connectors and MCP; a desktop agent such as MOOGH executes on your machine with your files and permissions.
- Do I need MCP to use it? No, but if your team already exposes tools over MCP, those tools become reachable without writing a bespoke connector.
- Does it replace human approvals? Not by design - the audit trail exists precisely because the agent is expected to act under defined oversight rather than unsupervised.
Related reading: