Google launches agentic Gemini for enterprises first

Google unveiled a unified agent inside Gemini at its Cloud event that can execute tasks on a user's behalf from a single interface, not just answer questions. Sundar Pichai said the platform now tops 1 billion monthly active users and that nearly 90% of the Fortune 100 already run Gemini Enterprise. The rollout targets businesses first; a consumer release will follow.

Goals, not instructions

Thomas Kurian, Google Cloud chief executive, described the agent as receiving "goals, not just instructions." It plans the work, invokes tailored skills and plugs into internal systems to finish the job. Model selection is automatic by default, though users can override it manually — starting with Anthropic's Claude models. Google pledged to extend the picker to open-source and additional proprietary models.

Its own Workspace identity

The agent provisions a dedicated Workspace account, effectively joining the org chart as a digital employee: separate email address, organizational context, awareness of teams, time zones, approval workflows and calendars. Colleagues invoke it by @-mention, email, file share or group chat. Every action writes to an audit trail credited to the agent, not the human. A "Task Box" surface exposes the reasoning chain, delegation to sub-agents, skill loading, code execution and progress.

Broad integration and cost controls

Connectors cover Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, Snowflake and more, with support for the Model Context Protocol inside and outside the corporate network. Clients run on iOS, Android, Windows, Mac, CLI, ServiceNow and Slack. Early testers included On, Shopify and PayPal; named enterprise customers include BNP Paribas, Merck and Orange Spain. Google added flexible spending options, multi-model orchestration, smart routing and real-time spend caps to keep AI costs in check.