SNACK: 3-line summary
- Google Cloud has announced a Gemini work agent designed to keep tasks running after you close your laptop.
- Google says it carries shared context into Workspace apps and can orchestrate work across Gemini and Claude models.
- The enterprise announcement does not promise access to every feature for every account.
Google Cloud has announced a Gemini agent for enterprise work that, according to the company, can keep a task running for hours or days after you close your laptop. The work runs in the cloud, with the same memory and context available across devices and channels.

Snackgirls react
AIKO: The laptop gets a break; the job’s context doesn’t. I’d like to see how it keeps a long task coherent when different steps use Gemini and Claude.
Red: I’d try it on a project pitch. Let it handle the research and numbers while I work out the story I want the slides to tell.
From market research to Sheets to a slide deck
Google says the agent works inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, carrying its memory, skills and controls with it. Its example follows one project through several applications: research market trends, build a financial model in Sheets, then create a slide deck without re-explaining the project at each step.
Teams can also create coworker agents with their own Workspace accounts, including an email address, calendar and Drive. Colleagues can add an agent to a Chat space or tag it in a document comment. It sees only the context shared with it, under the team’s existing sharing and membership permissions.
The Gemini agent can also use Claude
The Gemini name identifies the agent; the model doing the work is a separate choice. Google says it can orchestrate work across its Gemini models and Anthropic’s Claude models today, with support for other private and open models planned.
Google also announced Smart Routing, intended to match workloads to models based on performance and cost. This model choice sits beneath the agent’s working context, so using a different model does not require moving the team’s context, skills or data.
Agents get their own identities and access limits
Google says agents receive separate identities with least-privilege, role-based access, and their actions are recorded in audit logs. It describes an Agent Sandbox for task execution and an Agent Gateway that applies company policies to network traffic.
For a team delegating work, the practical question is which files the agent may see and which operations it may perform. Those permissions still need to be set deliberately; the announced controls do not remove every risk.
A spending cap can pause the job
Google says project-level spending caps track both token usage and sandbox costs. When a cap is reached, that project’s agent pauses, with an option to resume work from the Cloud Billing Console.
The announcement does not set out a single rollout date covering every account. Before handing the agent a workflow, check which announced capabilities your organization can access.
Sources and date: 2026-10-09

Comments
0No login needed. Edit or delete your comment from the same browser.
All comments 0
한국어 · English · 日本語No comments yet. Start the conversation.