Google Gemini Agent at Work: Preview, Features, Price and Access

Quick answer — October 8, 2026: Google Cloud has announced the Gemini agent, a universal workplace AI agent that can plan multi-step work, use company-approved tools, prepare documents, and run code. It can operate through Gmail, Drive, Docs, Sheets, Slides, Chat and Calendar, and connect to services such as Slack and Microsoft 365. The new agent is in an enterprise private preview, not a feature that every personal Gemini user can immediately switch on. Google has not announced a standalone public price or a general-release date for this exact new agent. That distinction is essential before buying a subscription.

The announcement arrived at Google’s Gemini at Work 2026 event. Google’s official product announcement describes one interface for assigning an outcome instead of supervising every step, while The Verge’s launch reporting specifies the private-preview access limitation. Here is a buyer-focused guide to what is real, what remains unavailable, and how an organization should evaluate it.

Gemini agent: the essential facts at a glance

What is different about Google’s new Gemini agent?

Most people know Gemini as a conversational assistant: enter a question, receive an answer, and then manually apply it somewhere else. The new agent aims to narrow the gap between answering and acting. For example, a manager could ask for a project status update, and the agent could locate information in permitted files, draft a presentation and put a finished artifact back into the team’s workflow. Google says the agent retains project context and can use specialized skills rather than starting every request as an isolated chat.

This remains an announcement about intended product functionality, not proof that every scenario will succeed without review. An accurate evaluation must separate Google’s demonstrated capabilities and customer examples from guaranteed outcomes in a buyer’s specific environment. An agent’s effectiveness still depends on permissions, connected apps, data quality and guardrails.

One interface across Workspace and other tools

Google lists Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar among the surfaces where Gemini can carry its context. It also describes connectors to organizational systems, with reporting identifying Microsoft 365 and Slack integrations. That matters for mixed-tool teams: an enterprise agent becomes more useful if it can reference the right records without copying sensitive files between personal accounts. It does not mean every third-party connector is enabled by default or allowed to access every folder.

Models are selected to suit the job

Google separates the agent from the underlying language model. According to its keynote write-up, Gemini can orchestrate work across Google’s model family and Anthropic Claude models, with additional models planned. Cost-aware routing is part of the pitch: a simple classification task should not necessarily use the same model as an involved planning exercise. Google’s architecture also introduces spend caps and governance controls. Ask your vendor representative which model choices and limits are actually enabled for your edition and preview.

What are coworker agents, and why do they have email addresses?

A notable feature is the coworker agent: a persistent digital team member with its own organizational identity rather than an invisible assistant borrowing an employee’s credentials. Google says a coworker can receive a company email account, calendar, Drive storage and a directory listing. Colleagues can mention it in Google Chat, assign it work and see its comments or edits under the agent’s identity.

For instance, a marketing team might ask an events coordinator agent to assemble launch-readiness notes from approved documents. The agent can draft an update and return it to the team, while employees remain responsible for checking the facts, audience and permissions before distribution. This is more auditable than an automation that silently signs in as a person, although separate identity alone does not eliminate security or hallucination risks.

Who can use Gemini agent now? Preview versus trial

Private preview is the key access restriction. As of October 8, coverage of Google’s announcement describes the new universal agent as available to selected enterprise customers. A Google Workspace subscription, a free consumer Gemini account, or even a regular Gemini Enterprise trial should not be assumed to include access to the newly announced preview. Google has not provided a universal self-service toggle or a firm date when all customers can use it.

There is an important product distinction: Google’s Gemini Enterprise app product page advertises a 30-day trial for certain editions, but that established product offering is not itself a promise that today’s new Gemini agent preview is included. Organizations should ask their Google Cloud account contact which exact features can be enabled for their tenant. Avoid websites demanding immediate payment for supposedly guaranteed preview access.

Practical steps for interested organizations

1. Check the official Gemini Enterprise page and existing Workspace edition before making a purchase decision.
2. Ask Google Cloud sales or your administrator whether the new Gemini agent at Work 2026 private preview is available to your tenant, rather than asking only about Gemini chat.
3. Specify two or three narrow workflows and the exact data sources the agent would need. Avoid beginning with blanket inbox or drive access.
4. Confirm administrator controls for connectors, agent identities, audit logs, approval stages and spending limits.
5. Measure output accuracy, staff review time and real costs before extending the pilot to financial or customer-sensitive actions.

How much does the Gemini agent cost?

Google did not give a separately confirmed retail price for this newly announced agent in its October 8 launch material. Existing Gemini Enterprise editions have their own commercial plans, trial terms, quotas and possible usage charges. Those are not interchangeable with an unreleased agent-specific price. A third-party price estimate for a different Gemini feature is not evidence of the price for this preview.

When comparing offerings, request an itemized statement covering base seats, model usage, connected-app requirements and any preview entitlements. Google advertises cost controls and spending caps, but a cap limits spending; it is not the same as free execution. The bottom line: there is no verified one-number price for universal Gemini agent access yet.

Gemini agent versus consumer Gemini and existing Gemini Enterprise

For a separate consumer-account change happening on October 9, see AVARIXO’s Gemini free-tier changes guide. Readers comparing integrations can also review our guide to ChatGPT custom MCP connectors; it addresses a different ecosystem, not an identical product.

Security questions an administrator should ask before enabling agents

Google highlights identity and policy management, agent sandboxing, network gateways, permission controls, usage records and spending caps. These features are valuable only when correctly configured. An administrator should test whether an agent can see a private folder it was not explicitly shared into; whether an unintended prompt inside an external document can redirect its behavior; whether sensitive actions require human approval; and whether logging captures which agent acted.

Start with low-risk, reversible work such as summarizing approved project notes. Do not use private-preview claims to justify unattended payments, employment decisions or disclosure of confidential records. Google’s October 8 launch recap makes clear that enterprise controls are central to the product design, not an optional afterthought.

Frequently asked questions

Is Google’s new Gemini agent free?

No free general-access entitlement has been confirmed for the new enterprise private-preview agent. The availability of a trial for the broader Gemini Enterprise app should not be mistaken for universal preview access.

Can I use it in Gmail and Google Docs?

Google says the agent can work within Gmail, Docs and other Workspace surfaces. Whether your organization can use that particular new agent today depends on its preview eligibility and administration settings.

Does it support Claude models?

Yes. Google’s official announcement says its model-routing architecture already supports Anthropic Claude alongside Google’s models.

Can it have an independent email address?

Google says coworker agents can receive dedicated Workspace identities, including email, calendar and Drive resources, with access scoped to shared organizational context.

When is the public release?

As of October 8, 2026, Google has not published a universal general-availability date. Treat precise dates on unofficial sites as predictions until Google confirms them.

Sources: Google Cloud launch keynote (October 8); Google launch recap; official product and trial details; The Verge reporting (October 8). Featured image: official Google Cloud announcement visual.