What is the Gemini agent? Features and pricing

What the Gemini agent is and what it does

The Gemini agent is Google Cloud's general-purpose AI agent for work: you describe the result you want, and it plans the steps, uses your company's tools and returns the finished work. Google Cloud's announcement puts answering questions, knowledge work, image and media creation, and writing and running code behind a single agent and a single API.

Announced at Gemini at Work 2026

Google Cloud CEO Thomas Kurian introduced the Gemini agent in his keynote at Gemini at Work 2026 on October 8, 2026 (US time). Google pitches it both as a personal assistant and as a member of a team, for example a project manager for a group or an analyst inside a finance department.

Google already offers Gemini Spark, an agent for individuals. The Gemini agent is the business version, connected to a company's own data, systems and controls.

Hand the Google AI agent objectives, not instructions

The main change from a chatbot is what you type: instead of walking the AI through each step, you state the outcome and come back later to collect it. The agent works out the order of tasks on its own.

Because it runs in the cloud, a job that takes hours or days keeps going after you close your laptop. Its memory and context follow you from your phone to your desktop. For jobs with many parts, it spins up short-lived sub-agents, each with its own identity, and runs them side by side or one after another.

Gemini features in Google Workspace

The Gemini agent's features fall into four groups: working inside Workspace apps, creating coworker agents, switching between Gemini and Claude models, and connecting to outside tools. Across all of them it keeps four kinds of memory, covering the task at hand, what it has learned from documents and people, how a job gets done, and what it has already done.

Gemini in Gmail, Docs and five more Workspace apps

In Workspace, the agent works inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, so there is no copying text out to a separate chat window. It carries the same memory, skills and admin settings into each of those seven apps. According to 9to5Google, you call it up by typing @Gemini in Gmail, Docs, Sheets, Slides or Chat.

Coworker agents with their own Workspace account

Describe a role, and Gemini builds a coworker agent to fill it. That agent gets its own identity, an address in the @agents.company.com format and its own storage. In Workspace it receives a full account with email, a calendar and Drive, and it shows up in the company directory.

Teammates work with it the way they would with a new hire: add it to a Chat space or @mention it with a request. In Google's example, an events coordinator agent drafts a launch readiness document and posts it back to the group. In Docs it suggests edits through comments, and its changes appear in version history under its own name.

Gemini and Claude models picked per job

Google separates the agent from the model underneath it, and the Gemini agent chooses for each job between Google's Gemini models and Anthropic's Claude models. Google says it will add other private and open models later.

The reasoning is cost and accuracy. A hard task goes to a stronger model, while routine work goes to a cheaper one. Google also argues that keeping the model swappable protects your context, skills and data when the leading model changes, which it says happens every few months.

Gemini MCP servers, skills and memory

The agent plugs into the software a company already runs. Google lists these connections:

• Collaboration: Confluence, Microsoft Office, Teams, Slack and Workspace
• Development: Git and Jira
• Business systems: Salesforce and ServiceNow
• Data: BigQuery, Databricks, Postgres and Snowflake
• Files stored on your own desktop

It can also reach any MCP (Model Context Protocol, an open standard for connecting AI to outside tools) server, whether inside or outside the company network. Repeatable procedures can be saved as skills. Teams publish skills to a company-wide registry, and individuals can keep private ones.

Use Gemini from Workspace, Microsoft 365 or Slack

You are not tied to Google's apps: the Gemini agent also takes requests from Microsoft 365, Slack and the command line. The table lists the devices and channels Google named.

The last row matters for developers. Without a screen of its own, the agent can sit behind an internal business app and do the work there.

For a company that lives in Microsoft 365 or Slack, this means a pilot does not require moving people to Workspace first. IT can test the agent in the tools staff already open every morning. Because the agent runs in the cloud, the memory and context it builds in one channel carry over to the others.

Gemini agent pricing and availability

As of October 2026, the Gemini agent is in private preview, and Google's keynote post gives no price and no launch date. Google is more specific about the industry editions: versions for financial services and legal are in preview, while government, healthcare and retail are listed as coming soon. The financial services edition comes with more than 50 foundational skills and draws on data from FactSet, LSEG, S&P Global and SEC filings.

Private preview, then select Gemini business plans

9to5Google reports that wider availability is coming soon for Workspace customers on select Business and Enterprise plans. Google has not said which plans those are or when the rollout reaches each region. Companies that already pay for Gemini Enterprise are best placed to ask their Google Cloud account team about preview access and timing.

No extra cost for Gemini Enterprise customers, Google says

Google told VentureBeat that Gemini Enterprise customers will get the Gemini agent at no additional cost. That includes companies with Gemini Enterprise through existing Workspace Business and Enterprise subscriptions.

What Google has not released is a plan-by-plan list of what each tier includes. Once general availability arrives, check your own contract against Google's published plan details.

AI agent security and Gemini cost controls

An agent that reads mail and edits files on people's behalf needs a clear record of who did what, and Google's answer is per-agent identity plus network and spending limits. Here is what an IT team can control.

Agent identity, permissions and audit trails

Every agent receives a cryptographically attested identity and, like an employee, only the permissions its job needs. Security administrators approve fine-grained, role-based access. When the agent connects to an outside system, its identity travels through standards such as OAuth.

Each action the agent takes lands in an audit trail under the agent's name, not a person's. If something goes wrong, the log shows which agent acted.

Agent Gateway and per-project Gemini spend caps

Agents run inside Agent Sandbox, and all of their traffic passes through Agent Gateway, which Google describes as a network firewall for AI. An admin writes a policy once, such as keeping agents out of documents classified "Need to Know", and it applies to every agent in the company.

On cost, Google says per-token prices have fallen 98% since 2024 while usage has surged, which makes budgets easy to overrun. In the Cloud Billing Console you can set a hard AI spending limit per project. When a project hits its cap, its agent pauses, and one click in the console resumes it.

Gemini agent vs Gemini Spark and Claude for Google Workspace

The Gemini agent is built for companies, while Gemini Spark is Google's agent for individuals. Claude for Google Workspace, from Anthropic, is a sidebar add-on that edits the Docs, Sheets or Slides file you have open.

Only the Gemini agent switches between Gemini and Claude models. The Gemini app's agent mode is a consumer feature with a similar name and a different audience. For more on Spark, see Gemini Spark for Mac: Beta, Download, Release Date.

The Claude add-on works on the one file you have open, while the Gemini agent carries a task across several apps. A team could use Claude for edits inside a single spreadsheet and the Gemini agent for work that spans email, documents and chat. The add-on is covered in Claude Google Sheets, Docs and Slides add-on guide.

Hand the Gemini agent whole jobs, not single prompts

The Gemini agent is Google Cloud's AI agent for work: hand it an objective and it returns the finished job. It works inside seven Workspace apps, creates coworker agents with their own email address, and picks Gemini or Claude models per task. Admins assign permissions per agent and cap AI spending per project.

As of October 2026 it is in private preview, with no public price and no regional launch date. Google told VentureBeat that Gemini Enterprise customers will not pay extra, so the first step is to check whether your current plan is on the list once Google publishes it.