Google Cloud launches Gemini agent, one agent for all office work

What's new

On October 8, 2026, at the Gemini at Work event, Google Cloud showed a new "Gemini agent." Google calls it a universal agent for work. You write one task, and the agent plans the steps itself, uses the necessary tools and delivers a finished result. It can answer questions, do knowledge work, create images and media, and write and run code.

The agent works in Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. According to media reports, it can also be used via Microsoft 365 and Slack. At the same time, special editions are coming for specific industries. The versions for finance and legal can be tried in preview now. The versions for the public sector, healthcare and retail are coming soon.

What's clever

The agent is not just a chatbot. For larger tasks, it creates sub-agents on its own, small helpers that each take a part. It also chooses which AI model fits the task, and it can keep running in the cloud even if your device is turned off.

It can be connected to the company's own systems, such as Salesforce, ServiceNow, Jira and Snowflake. It also works with any MCP server, a kind of standard plug between AI and programs. The finance agent uses data from FactSet, LSEG and SEC filings, among others, for investment analysis, credit assessment and risk models.

Security is a big selling point. Each agent gets its own identity with as few permissions as possible, and everything is logged. It runs in an isolated sandbox, and an "Agent Gateway" acts as an AI firewall that enforces the company's rules in real time.

Cheaper or better?

The price has not been disclosed yet, nor has it been disclosed when the agent will be available to everyone. Google only mentions built-in cost limits. You can set spending caps in real time via the Cloud Billing Console, and the choice of model per task is meant to keep the bill down.

Google has also not shown benchmarks comparing the agent directly with competitors, so we cannot say whether it is better than offerings from Microsoft, OpenAI or Anthropic. Anthropic's Claude models are mentioned as an option in the platform. Google's argument is rather the breadth. One agent works across many systems, and it is available in both Google's and Microsoft's tools.

The numbers show that demand is high. According to Google's CEO Thomas Kurian, nearly 80 percent of Google Cloud's customers use the company's AI products. Nearly 90 percent of the Fortune 100 work with Gemini Enterprise. About 500 customers have each processed over a trillion tokens, that is, pieces of text, in the past year. Google Cloud's revenue grew 82 percent in the second quarter of 2026.

What it's good for

The agent is aimed at businesses, not private users. It suits heavy, repetitive tasks across multiple systems, such as gathering research, making reports and presentations, writing code or reviewing data and documents.

Finance departments and law firms can test the industry-specific editions right now. For Danish companies, it is important to wait for information about price and availability in the EU. It has also not been disclosed whether the agent can be used in Denmark from the start. The control features are interesting for those who are nervous about a self-driving AI getting too free a rein.