Google Cloud Launches Gemini Agent for Enterprise Work

Google Cloud used its Gemini at Work 2026 event today to introduce the Gemini agent, a single, universal agent for work. Rather than juggling a separate assistant for every task, employees hand over an objective and the agent plans the work, calls the right tools, and comes back with something finished inside the apps they already use.

What happened

In Thomas Kurian's keynote post, the Google Cloud CEO framed the shift simply: "You give it objectives, not instructions. You delegate an outcome and come back to finished work." The agent runs across web, iOS, Android, Windows, Mac and the command line, and it lives inline in Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, plus Microsoft 365 and Slack. No dedicated interface required.

The architectural detail that stands out is identity. A Gemini agent configured as a digital coworker gets its own Workspace account, email address, calendar, Drive storage and company-directory presence, so managers can assign it work the same way they assign it to a colleague. Sessions persist for hours or days, survive device switches, and can spawn specialised sub-agents for complex jobs.

Model choice is deliberately open. Small tasks route to Gemini Flash, long-horizon work to a frontier model, and Anthropic's Claude models are available today with more proprietary and open-weight models promised later. Governance includes cryptographically attested agent identity, audit logs, an Agent Gateway for external calls, and sandboxing for code execution. Coworker agents only see context that team members explicitly share, not a human's full permission set.

Why it matters

Google is not shipping this into a vacuum. Microsoft refreshed Copilot on September 25, OpenAI introduced its always-on dots agents days later, and Anthropic wired Claude into Google Docs, Sheets and Slides on October 6. Every major lab is racing to turn the assistant into a long-running digital worker with its own identity and memory.

For enterprises, the practical differentiator is the cost layer. Google pairs multi-model orchestration and Smart Routing with hard spend caps in the Cloud Billing console. When a project hits its cap, the agent pauses instead of quietly burning budget, and tracking is per project so AI spend can be charged back to specific departments.

What this means for builders

If you build automations, the interesting part is not the chat window, it is the plumbing. Persistent agent identity, a shared skill registry, and policy enforcement through a gateway are the primitives that make agents auditable. Google also ships reusable skills that turn a one-off task into a template. That is the same pattern n8n builders already use with reusable workflows, now arriving as a managed enterprise service.

Watch the open questions too. Google has not published how provisioning, licensing or per-component permissions will work, and it has not said exactly when every capability reaches customers. The agent platform race is now as much about governance and billing as it is about model quality.

Sources: Google Cloud blog · Google blog · SiliconANGLE · VentureBeat