Google announced on Thursday a unified Gemini agent that can plan work, use tools, and act on users' behalf from a single interface — and it starts with businesses, where nearly 90% of Fortune 100 companies already run Gemini Enterprise.

CEO Sundar Pichai opened the Google Cloud event with the scale argument: Gemini has over 1 billion monthly active users. Corporate adoption is the distribution channel. Google will bring the agent to enterprises first, consumers later, because — as Pichai put it — launching powerful agents requires solving the "harder problems around security, scale, and performance."

What can the agent actually do?

Google Cloud CEO Thomas Kurian said the new agent can be given "objectives, not just instructions." In practice, that breaks down to:

• Planning work and delegating tasks to subagents
• Using custom skills and connecting to internal systems
• Handling requests with attachments — files, folders, project workstreams
• Working with any Model Context Protocol (MCP) server, inside or outside the corporate network

Integrations named at launch include Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, and Snowflake.

Who picks the model?

By default, the AI selects the best model for the task. Users can override that choice — including with third-party models, starting with Anthropic's Claude. Google said the model picker will expand to open source and other private models later.

How does oversight work?

A "tasks inbox" shows Gemini's thinking process, subagent delegation, skill loading, code, and progress. The agent also gets its own Workspace account with its own email address and context — effectively another co-worker. It knows who sits on which team, their time zones, who approves items, and what is on calendars. Users can tag it, email it, share with it, or add it to group chats. Every action writes an audit trail attributed to the agent, not to a person.

Access spans iOS, Android, Windows, Mac, the command line, Google Workspace, Microsoft 365, ServiceNow, and Slack.

What does this change for cost control?

Google paired the launch with flexible spending options — multi-model orchestration, smart routing, and real-time spend caps — aimed at keeping enterprise AI bills predictable.

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

The third-party model picker and the spend caps signal where this goes next: Google is positioning Gemini less as a chatbot and more as orchestration infrastructure, with the consumer rollout to follow once the enterprise security and scale problems are solved.