Google turns Gemini into a working agent with its own email

By Daniel Okafor

Google announced on Thursday a unified Gemini agent that does not just answer questions but completes work on the user's behalf, launched first for businesses before consumers.

The debut lands as AI tools shift from chat to action. Competing agents are already booking travel, writing code and scheduling meetings: Meta's Muse, messaging agents like Instinct, and ChatGPT's recent Dots.

Google starts with two numbers on its side. CEO Sundar Pichai told the Cloud event that Gemini has over 1 billion monthly active users. He said nearly 90% of Fortune 100 businesses already run Gemini Enterprise at work.

Why businesses first?

Corporate adoption explains the sequencing. Google will ship the agent to enterprises before consumers. Pichai said the approach lets the company solve the "harder problems around security, scale, and performance" that powerful agents create.

Thomas Kurian, CEO of Google Cloud, framed the capability in one line: the agent can be given "objectives, not just instructions." It plans the work, uses custom skills and tools, and connects to internal systems to finish the job.

How does the agent actually work?

The mechanism is a shift from question-answering to task ownership. By default, the AI picks the best model for each task. Users can override that choice and select third-party models — starting with Anthropic's Claude. Google said the model picker will later add open source and other private models.

Requests can carry attachments: files, folders, or project bundles that combine files and skills for specific workstreams. The agent plugs into business data and systems, including:

• Google Workspace and Microsoft 365
• Slack, Jira and Confluence
• Git, BigQuery, Databricks, Postgres and Snowflake

It also connects securely to any Model Context Protocol (MCP) server, inside or outside the company network.

Users track progress through a "tasks inbox." There they can see Gemini's thinking process, its delegation of tasks to subagents, the loading of special skills and code, and overall progress.

What does it mean to give an agent a desk?

The most concrete design choice: the agent gets its own Workspace account, treated like another co-worker. It has its own email address and its own context, Google said. It knows who sits on which team, their time zones, who approves items, and what is on people's calendars.

Users reach it by tagging, emailing, sharing files with it, or adding it to a group chat. Every action it takes writes an audit trail attributed to the agent, not to a person.

The agent runs on iOS and Android devices, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365, ServiceNow, and Slack.

Who is testing it — and what does it cost?

Early testers include sportswear brand On, Shopify, and PayPal. Google named Gemini Enterprise customers BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers.

Cost control is part of the pitch. Google said flexible spending options — multi-model orchestration, smart routing, and real-time spend caps — will help businesses keep enterprise AI costs in check.

The enterprise-first route gives Google time to harden security before a consumer launch, and the model picker with Claude inside a Google product signals that the agentic market will be judged on task completion, not model loyalty.