Gemini Agent: Google Cloud's Universal Work Agent

Introduction

On October 8, 2026, at its Gemini at Work 2026 event, Google Cloud announced the Gemini agent, which it calls "a single, universal agent for work." In a keynote-adapted post, Google Cloud CEO Thomas Kurian says 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." Google's framing is that "work now starts in the prompt window."

This is a business-focused product. According to The Verge, the agent is currently available only to enterprise customers in private preview, inside the Gemini Enterprise app. TechCrunch reports that Google will focus on businesses first and roll the agent out to consumers later. It is separate from the consumer Gemini Spark agent that Google launched in May (see our earlier Gemini Spark coverage), which The Verge notes lives in a separate interface in the Gemini app.

Feature Overview

1. One agent, many surfaces. Google's post lists web, iOS and Android, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365 and Slack as access points, and says the agent can run headless inside third-party applications. It runs in the cloud, so Google says memory and context persist across devices and work that takes hours or days keeps running after a laptop is closed.

2. Sub-agents and coworker agents. Google says the agent can create temporary, job-specific sub-agents, each with its own identity, and can act as a "coworker agent" with a persistent role. Coworker agents get dedicated identities, including their own @agents.company.com email addresses and persistent storage, and only see context that users or teams provide. Inside Workspace, a coworker agent can receive an account with email, calendar and Drive, and can be added to a Chat space or @mentioned. Edits it makes appear under its own name in version history.

3. Tools, skills and memory. The agent connects to systems such as Confluence, Microsoft Office, Teams, Slack, Workspace, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake, and to any Model Context Protocol (MCP) server inside or outside the company network. Teams can publish tools and skills to shared registries. Google describes four memory types: session, semantic, procedural and episodic.

4. Model choice. Google says the agent orchestrates across its own Gemini family and Claude models from Anthropic today, with other private and open models planned. TechCrunch adds that by default the agent picks the model, and users can override the choice.

5. Data and industry skills. For data teams, Google describes skills that generate PySpark code, provide notebooks, train models and fix pipeline issues, plus reporting skills tied to BigQuery and a Knowledge Catalog. Google says Bloomberg Media lifted SQL query accuracy by 63% during initial development after grounding its data agents in the Knowledge Catalog. Industry specialization is "in preview for Financial Services and Legal, and coming soon to Government, Healthcare, and Retail." Google says the Financial Services offering has more than 50 foundational skills and draws on FactSet, LSEG, S&P Global and SEC filings.

6. Governance and cost controls. Google says every agent has a cryptographically attested identity, actions are written to an audit trail attributed to the agent, and agents run in an Agent Sandbox with traffic passing through an Agent Gateway that enforces organization policy. For spending, Google lists multi-model orchestration, smart routing and real-time spend caps; when a project's cap is triggered, its agent pauses and can be resumed from the Cloud Billing Console.

Usability Analysis

TechCrunch describes a "tasks inbox" where users can follow the agent's reasoning, delegation to sub-agents, skill loading, code and progress, and says the agent can be summoned by tagging it, emailing it, sharing with it or adding it to a group chat. Google's post gives workflow examples such as scheduling a meeting with a team's regional event leads without supplying names, and one-click delegation of a slide-deck request found in an email.

TechCrunch names On, Shopify and PayPal as early testers. Google's own customer examples mostly concern Gemini Enterprise broadly rather than the new agent, so they should be read as context on adoption, not agent results. Google says nearly 90% of the Fortune 100 use Gemini Enterprise, and TechCrunch's event report attributes to Sundar Pichai a figure of over 1 billion monthly active users for Gemini.

Pros and Cons

On the plus side, the design addresses issues enterprises raise often: identity, audit trails, sandboxing and spend caps. Support for MCP, multiple third-party systems and Claude models reduces lock-in concerns. On the minus side, access is limited to a private preview, the industry packages are still in preview, and the keynote post is vendor-authored, so performance claims have not been independently tested.

Outlook

Google says it chose to start with enterprises to tackle what Pichai called "harder problems around security, scale, and performance," per TechCrunch. The product's value will depend on how well the cloud-persistent agent, coworker identities and policy tooling work in real deployments, and on how the preview broadens. Google also says the model picker will later expand to open source and private models.

Conclusion

The Gemini agent consolidates chat, delegated work and code generation into one governed agent for organizations that already use Google Workspace or Gemini Enterprise. It is best suited to IT, data and operations leaders evaluating enterprise agent platforms, and less relevant to individuals for now. Rating: 4 out of 5.

Editor's Verdict

Gemini Agent: Google Cloud's Universal Work Agent earns a solid recommendation within the Gemini space.

The strongest case for paying attention: one agent spans chat, delegated work and code generation across Workspace, Microsoft 365, Slack and the command line. That alone raises the bar for what readers should expect in this space. Reinforcing that, Google says the agent can use Claude models alongside Gemini, which reduces dependence on a single model — practical value rather than just headline appeal. The broader signal worth registering is straightforward: a single agent replacing separate chat, automation and coding surfaces is the central pitch, with Google's post saying work now starts in the prompt window. On the other side of the ledger, one constraint is real rather than a marketing footnote: access is limited to enterprise customers in a private preview, according to The Verge, so most individuals and small teams cannot try it yet. It should factor into any serious decision. Layered on top of that, industry specializations are in preview for Financial Services and Legal, with Government, Healthcare and Retail still listed as coming soon — which narrows the set of teams for whom this is an obvious yes.

For IT, data and operations leaders at organizations already on Google Workspace or Gemini Enterprise, this is a serious evaluation candidate, not just a curiosity to bookmark. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.