Tram cong nghe AI
The Gemini agent is Google’s new “universal” AI agent for work, introduced by Google Cloud CEO Thomas Kurian at the Gemini at Work event on October 8, 2026. It combines the chat, the autonomous worker and the coding agent into one window. The surprise: you can run it on Google’s own models or on Anthropic’s Claude models, and companies can set a hard spending cap per project.
Key Takeaways
- Launch: it was announced October 8, 2026, at Gemini at Work.
- One window for everything: chat, autonomous tasks and coding in a single prompt box. “Work now starts in the prompt window,” said Kurian.
- Model choice: Google’s Gemini models and Anthropic’s Claude models today, with other models promised later.
- Works everywhere: web, Android, iOS, desktop and command line, plus inside Google Workspace, Microsoft 365 and Slack.
- Cost controls: automatic routing to cheaper models, hard spending caps per project, and per-project billing.
- Price: no additional charge where Gemini Enterprise is available, according to a Google spokesperson. Only a limited number of companies have access so far.
What Is the Gemini Agent?
It is an AI agent that takes an objective, not step-by-step instructions, and works out the steps itself. Google wants working with agents to feel as intuitive as Google Search did when it launched.
Google shared two examples:
- Ask it to set up a meeting with “the usual team of regional event leads.” The agent works out who they are from your chat space and past email threads, checks their calendars and starts an email thread to find a time.
- Ask it to research market trends, build a financial model in Sheets and turn both into a slide deck, without re-explaining the project at each step.
The agent runs in Google’s cloud, so it keeps a persistent memory no matter which device you use to reach it.
Gemini Agent Features
| Feature | What it does |
|---|---|
| Universal prompt window | Chat, autonomous work and coding in one place |
| Helper agents | For big jobs, it creates temporary helper agents, each with one task and its own identity for auditing |
| Coworker mode | A standing team role (for example, event coordinator) with its own Gmail address, calendar and file storage |
| Ready-made skills | Ships with built-in skills and four kinds of memory |
| Connections | Admins choose which enterprise tools it can use, or add more through MCP (Model Context Protocol, a standard way to plug tools into AI agents) |
The Gemini agent as a coworker
In coworker mode, teammates can add it to a chat or tag it like a colleague. It appears under its own name in a document’s edit history. It only sees what team members share with it.
Gemini Agent Models: Gemini and Claude
Google separates the agent from the AI model underneath it. Right now, the choice is limited to Google’s models and Anthropic’s Claude models, and Google says other models will follow.
You can pick a model yourself, or let the agent choose. It picks a model based on the job, aiming for accuracy while keeping costs down. Google says this lets companies switch models later without losing their data or settings.
This is a notable strategy. Google’s newest flagship, Gemini 4 Argon, improved its standing, but by sitting on top of any model, Google can still be “where the work gets done” even when a rival has the best model. If you already use Anthropic’s models, see our Claude AI review.
How to Control Gemini Agent Costs
Google acknowledged that AI budgets are under strain as usage rises, even while per-token prices fall. It offers three ways to keep spending in check:
| Control | How it works |
|---|---|
| Automatic routing | Simple work is sent to cheaper models |
| Hard spending cap | Set a cap per project. When it is reached, the agent pauses until someone clicks to resume |
| Per-project tracking | Spending is tracked by project, so costs can be billed back to departments |
Practical setup tip: give each team or workflow its own project, set a cap on each, and leave model choice on automatic for routine tasks. That way, a runaway task pauses instead of draining the budget, and every department sees its own bill.
Gemini Agent Security: Identity and the Agent Gateway
Google says any company deploying agents should be able to answer four questions: Who is the agent? What is it allowed to do? What did it do? What should it never touch?
- Own identity: every agent gets its own identity and limited access based on its role.
- Full log: every action is logged under the agent’s name, not a person’s.
- Signs in as itself: when it connects to another company’s software, it uses the same login system behind “Sign in with Google.”
- Agent Gateway: agents run in a walled-off space, and all traffic passes through a checkpoint Google calls an “AI network firewall.” A rule written once, such as “agents may not open documents marked Need to Know,” applies to every agent.
Gemini Agent Pricing and Access
A Google spokesperson told VentureBeat the agent will be built into Gemini Enterprise and available “at no additional charge for customers where Gemini Enterprise is available, including through existing Google Workspace business and enterprise subscriptions.”
For now, only a limited number of companies have access. Google has not given a date for wider availability in the source we used. For a look at Google’s consumer-side agent features, see our post on Gemini Spark.
Is the Gemini Agent Worth It?
Pros
- One agent for chat, autonomous work and coding.
- Model choice, including Claude, without losing data or settings when you switch.
- Strong admin controls: agent identities, company-wide rules and spending caps.
- Works inside Microsoft 365 and Slack, not just Google Workspace.
- No extra charge for Gemini Enterprise customers.
Cons
- Access is limited to a small number of companies today.
- It is built for businesses on Gemini Enterprise, not individual users.
- Model choice is limited to Google and Anthropic for now.
- Google is late to the agent market; analysts say the real test is wide production use.
The verdict: if your company already pays for Gemini Enterprise or Workspace business plans, it is worth watching closely. The cost and identity controls are its strongest selling points.
Gemini Agent FAQ
What is the Gemini agent?
It is Google Cloud’s universal AI agent for work, announced October 8, 2026. It combines chat, autonomous work and coding in one window.
Does the Gemini agent use Claude?
Yes. You can choose Google’s Gemini models or Anthropic’s Claude models, or let the agent pick. Google says more models will follow.
How much does the Gemini agent cost?
According to a Google spokesperson, it comes at no additional charge for Gemini Enterprise customers, including existing Workspace business and enterprise subscriptions.
Can I limit Gemini agent spending?
Yes. Companies can set a hard spending cap per project. When the cap is hit, the agent pauses until someone resumes it.
Does the Gemini agent work outside Google Workspace?
Yes. It works inside Microsoft 365 and Slack, and you can reach it on the web, Android, iOS, desktop and command line.
Sources
Last updated: October 15, 2026
