Google Cloud’s Gemini Agent: One Agent for Work, With Industry Versions in Preview

Google Cloud has announced the Gemini agent, which it calls “a single, universal agent for work” [1][2]. Thomas Kurian, Google Cloud’s chief executive, unveiled it on 8 October at Google’s Gemini at Work 2026 event. The versions built for financial services and legal work are in preview, and Google’s announcement gives no general-availability date [1].

What Google says it does

According to Google, the agent answers questions, handles knowledge work, creates images and media, and writes and runs code, “all in a single agent and a single API” [1]. Users give it objectives rather than step-by-step instructions. It runs in the cloud and keeps “a single set of memories, context, and one personalization graph” across devices, so long jobs continue after a laptop is closed [1].

Two features stand out:

• Sub-agents. Gemini can create “temporary, job-specific agents, each with their own identity” to split up multi-step work [1].
• Coworker agents. These are longer-lived agents with a defined role. Google says each “receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory”. Colleagues can @mention it like a person, and it “sees only what you share with it” [1].

Google says the agent picks the model for each job from “our Gemini family of models and Claude models from Anthropic” [1].

How Google says it is fenced in

Giving software its own email account and the power to run code raises obvious security questions. Google’s answer, in its own description [1]:

• Identity. “Every agent gets its own identity, cryptographically attested”, with least-privilege permissions.
• Audit. Every action is logged and “attributed to the agent rather than to a person”.
• Sandbox and firewall. “All Gemini agents execute tasks safely inside an Agent Sandbox with its own network boundary.” All traffic passes through “Agent Gateway, an AI network firewall” that applies company policies.
• Spending. Real-time spend caps pause a project’s agent when a limit is reached.

These are Google’s design claims. TSN has not tested them, and no independent assessment has been published.

Who can use it

Google says the industry versions are “now in preview for Financial Services and Legal, and coming soon to Government, Healthcare, and Retail” [1]. Preview means a limited, pre-release stage: these versions are not generally available, and the announcement gives no date for wider release [1]. Google’s figures on adoption, such as “nearly 90% of the Fortune 100 use Gemini Enterprise”, are its own claims [1].

What this does not prove

• That the security controls work as described. Identity, sandboxing and the gateway are Google’s descriptions, not independently tested [1].
• That the industry versions are ready for general use. Finance and Legal are in preview; the others are “coming soon” [1].
• That an agent with its own accounts is low-risk. Coworker agents act under their own identity with whatever access people share with them. Mistakes or manipulation would act through that access.

The Bottom Line

Google is pitching one agent that can do most office and coding work, spin up helpers with their own identities, and sit in a team like a colleague, wrapped in Google’s own identity, sandbox and firewall controls. The industry versions are in preview only, and how well the safeguards hold up remains to be shown.

Sources

1. Thomas Kurian, “Gemini at Work 2026: Introducing Gemini agent”, Google Cloud Blog, 8 October 2026 (adapted from the keynote; company announcement). https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026
2. Google, “Google Cloud launches Gemini agent”, The Keyword (blog.google), 8 October 2026 (short company summary). https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/gemini-at-work/