Google Gemini Agent Tutorial: Set Up Your Universal AI Agent for Work

Introduction

On October 8, 2026, Google Cloud CEO Thomas Kurian opened Gemini at Work 2026 by announcing the Gemini agent, a single agent meant to replace the pile of separate AI assistants most teams now juggle. This tutorial explains what the Gemini agent is, who can use it today, and how to set it up for your team, including the coworker agents that get their own Workspace accounts. It is written for Workspace admins and team leads who need to plan a rollout, not just try a demo.

Quick answer

The Google Gemini agent is a single, universal agent for work that answers questions, handles knowledge work, creates media, and writes code from one prompt box. It is in private preview since October 8, 2026.

Evidence

Google is pitching the Gemini agent as the front door to all of its enterprise AI, and the scale numbers in the launch announcement back up why it matters. In the last year, nearly 500 Google Cloud customers each processed more than one trillion tokens, and nearly 90% of the Fortune 100 use Gemini Enterprise, according to Google’s Gemini at Work 2026 announcement. Those are the customers the agent is built for first, which explains the private preview rollout and the Workspace-first design.

The customer results Google cites are concrete rather than vague. Bradesco, one of Brazil’s largest banks, cut document review time from 1 hour to 5 minutes and reduced risk inconsistencies by 60% after deploying Gemini Enterprise. At Bunnings, an internal agent saved staff half a million hours of administrative work. PayPal routes 10 million multi-model requests every week through the same multi-model approach the agent now uses.

The core idea is simple, and Google states it in one line: “You give it objectives, not instructions.” You delegate an outcome, such as a finished launch readiness document or a reconciled budget model, and the agent plans the steps, uses your tools, and brings back finished work. That is a different contract from a chatbot, and it is why the setup work in this tutorial focuses on context, permissions, and guardrails rather than prompt tricks.

Prerequisites

• A Google Workspace organization on a Business or Enterprise plan, or a Gemini Enterprise subscription. The agent is built into Gemini Enterprise, and Google says wider availability will reach select Workspace Business and Enterprise plans first.
• Workspace admin access, or a direct line to your admin. Enabling the agent, approving coworker agents, and setting spend caps are admin decisions.
• A Google Cloud account contact, if your organization wants to ask about the private preview. Google has not published a self-serve signup for the preview.
• A clear first use case. Pick one repetitive, multi-step job with a visible owner, such as weekly reporting or meeting preparation, before you create any coworker agent.
• Basic familiarity with Google Workspace sharing. Coworker agents only see what you share with them, so your Drive folder structure and sharing habits decide what they can do.

Step-by-step guide

Step 1. Confirm whether your organization can get the agent yet

The Gemini agent is in private preview, so the honest first step is an eligibility check, not an install. There is nothing to download and no separate app to deploy. If your organization is a Gemini Enterprise customer, ask your Google Cloud account team whether your organization is in the preview group. If you run Workspace without Gemini Enterprise, ask your Workspace admin to check whether your plan is among the select Business and Enterprise plans Google says will get wider availability soon. Press reports at launch said Google expects wider availability to start in North America, but Google has published no general availability date, so treat any date a salesperson quotes as unconfirmed until Google posts it.

Step 2. While you wait, prepare the context the agent will need

An agent is only as good as the context it can reach, and Google says the Gemini agent “arrives already knowing your tools, data, and work history.” Use the waiting time to make that true. Organize the Drive folders your first use case touches, fix broken sharing on the documents the agent will need, and retire stale files that would confuse it. Write down the recurring jobs you want to delegate, with their inputs, their outputs, and who checks the result.

Step 3. Learn the three ways the agent works inside Workspace

Google describes three modes, and they need different setup decisions. Personal assistance is the agent working for one person, already briefed on their calendar, team, and projects. Proactive delegation is the agent suggesting tasks it can take, for example offering to turn a manager’s email asking for a project update into a slide deck with one click. The third mode is the coworker agent, a persistent agent with a defined role that works with a whole team. Start with personal assistance for individuals, add proactive delegation once people trust the results, and create coworker agents only for roles that genuinely repeat every week.

Step 4. Create your first coworker agent by describing its role

This is the step that makes the Gemini agent different from every chatbot your team has tried. You describe the role you need, and Gemini creates the agent. Google’s announcement puts it plainly: “The agent receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory.” The agent’s email address sits on your company’s agents subdomain, in the form @agents.company.com, and colleagues work with it the way they work with anyone else, by adding it to a Chat space or @mentioning it.

Give the role a narrow job description with a named owner. A good first coworker agent is an events coordinator that drafts launch readiness documents, or a reporting analyst that assembles the weekly numbers. A bad first coworker agent is a vague “help the team with everything” role with no owner and no definition of done. Name the person who reviews its output, decide which Drive folders it may access, and share exactly those folders with its Workspace account. Remember the access rule from the announcement: the agent “only have access to the context that you or your team members provide.”

Step 5. Connect the agent to your tools and teach it your skills

The agent connects to the software your company already runs. Google lists Confluence, Microsoft Office and Teams, Slack, and Workspace for collaboration, Git and Jira for development, Salesforce and ServiceNow for enterprise platforms, BigQuery, Databricks, Postgres, and Snowflake for data, and files on your own desktop. It can also connect to any Model Context Protocol (MCP) server inside or outside your company network, and teams can publish their own tools to an enterprise tools registry.

Skills are the other half of this step. A skill is a reusable set of instructions that teaches the agent how to perform a specific multi-step task, and Gemini ships with a global library of them. Build one skill for your first use case, then publish it with our guide to sharing Gemini skills with your team through a team registry. If your team still uses Gems, read our guide to migrating Gemini Gems to Skills first, because Gems are being replaced by skills starting in November 2026 for personal accounts.

Step 6. Give it an objective, not instructions

Once the coworker agent exists, delegate the way Google intends: state the outcome and let the agent plan. Instead of “open the spreadsheet, copy the numbers, and make slides,” try “produce the weekly launch readiness update as a slide deck from the tracker in Drive, flag anything that slipped since last week, and post it in the launch Chat space by Friday 9 a.m.” The agent can spin up temporary sub-agents for multi-step work, each with its own identity, and coordinate parallel and sequential steps that can run for hours or days. It keeps working in the cloud after you close your laptop, and the finished work is waiting when you come back. For long jobs, our tutorial on running multi-day persistent tasks with the Gemini agent and subagents covers how to structure work that outlasts a single session.

Step 7. Set the guardrails before the agent touches real data

Google built the governance layer into the announcement, and you should configure it before the first real delegation. Set spend caps per project so the agent pauses when it triggers them. Route traffic through the Agent Gateway, the policy firewall Google describes for agent traffic, and write one rule for the boundary that matters most, such as “agents may not open documents classified Need to Know.” Check that audit trails attribute work to the agent’s own identity, so a coworker agent’s edits show under its own name in version history. If your organization needs agentic OAuth for MCP connections, our guide to enabling Google’s agentic OAuth setting for Workspace MCP covers that setting before the November deadline. In regulated industries, look at the industry specializations too: our overview of the Gemini industry agents for financial services and legal explains what the preview versions add over the generic agent.

Troubleshooting

The @Gemini mention does not appear in Gmail, Docs, or Chat

This is the most common early problem, and it almost always means access, not a bug. The agent is in private preview, and press reports say it reaches Workspace customers on select Business and Enterprise plans first. Confirm with your Workspace admin that Gemini is enabled for your organization, confirm your plan is eligible, and ask your Google Cloud account team whether your organization is in the preview. If any of those three answers is no, no setting on your own account will make @Gemini appear.

A coworker agent says it cannot find a document you know exists

A coworker agent only sees what you share with it, and its access follows the sharing and membership your team already uses. Open Drive, check whether the document or its parent folder is shared with the agent’s Workspace account, and share it explicitly if it is not. Also check that the file is not sitting in someone’s personal Drive under sharing rules the agent cannot inherit. If the agent can see the folder but not one file inside it, that file almost always has different sharing settings from its parent.

The agent stops halfway through a long job

Two causes cover most cases. First, a spend cap may have triggered and paused the agent, which is the intended behavior. Ask your admin to check the project’s cap and raise it deliberately, not by disabling caps entirely. Second, the job may genuinely need input it cannot get, such as a decision only a human can make or a system it has no connector for. Check the agent’s status or summary first: it should say what it finished and what it is waiting on. Break the remaining work into a smaller objective and delegate that.

You cannot tell who changed a document, you or the agent

This is a design feature, not a defect. A coworker agent’s edits appear under its own name in version history, and its audit trail is attributed to its own identity. If you see changes you cannot attribute, open the document’s version history and look for the agent’s name rather than a colleague’s. If your team needs stricter separation, give each coworker agent a distinct role name so its version history entries are unambiguous.

You are confusing the Gemini agent with Gemini Spark

The names overlap and the products do not. The Gemini agent is the work product for Gemini Enterprise and Workspace accounts, with company context, admin controls, and coworker agents. Gemini Spark is the personal agent inside the Gemini app for consumer accounts on Google AI Pro or Ultra. If a feature you read about mentions a personal account, a consumer subscription, or the Gemini app, it is almost certainly Spark, and it will not appear in your Workspace admin console.

FAQ

Is the Google Gemini agent available to use now?

Not for most organizations. Google launched the Gemini agent in private preview on October 8, 2026, for a select group of customers. Wider availability is promised soon for select Google Workspace Business and Enterprise plans, but Google has published no general availability date as of November 6, 2026.

How much does the Gemini agent cost?

Google has published no separate price for the agent. Press coverage of the launch reports that Google intends to include it at no additional charge where Gemini Enterprise is available, and that it does not intend to charge a per-seat license for coworker agents in Workspace. Metering details are expected closer to wider availability.

How is the Gemini agent different from Gemini Spark?

The Gemini agent is the work product for Gemini Enterprise and Google Workspace accounts. Gemini Spark is the personal agent inside the Gemini app for consumer Google accounts on Google AI Pro or Ultra. They share the Gemini name but serve different accounts, data, and controls.

Can a coworker agent read all of my email and files?

No. A coworker agent only has access to the context you or your team members provide, and its access follows the sharing and membership your team already uses. Share a document or folder with the agent’s Workspace account and it can work with it. Do not share it, and the agent cannot see it.

Which AI models does the Gemini agent use?

Google routes each job to the model that fits best, orchestrating across its own Gemini family of models and Claude models from Anthropic today. Google says other leading private and open models will follow, so the model underneath the agent can change without moving your context, skills, or data.

Does the Gemini agent work outside Google Workspace?

Yes. Google says it runs on the web, iOS and Android, Windows and Mac desktops, and the command line, inside Google Workspace, Microsoft 365, and Slack, and as a headless agent embedded in third-party apps. Inside Workspace it works directly in Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar.

Conclusion

The Gemini agent is Google’s bet that work will start in a prompt box, with one agent that chats, works for hours or days, and writes code, all under the same memory, skills, and controls. Most organizations cannot use it yet, because it is still in private preview, but the setup work is not blocked: pick one repeatable job, organize the context it needs, build one skill for it, and agree the guardrails with your admin. As a next step, read our complete n8n guide if you want to compare Google’s managed agent with a self-hosted automation platform you can run today, or continue with the multi-day task tutorial linked above to plan work that runs longer than a single session.

Frequently asked questions