What Is Gemini agent? Features and Availability of Google Cloud's AI Agent for Work | CoAI

On October 8, 2026, at its enterprise event Gemini at Work 2026, Google Cloud announced Gemini agent, an AI agent for work. Google positions it as "a single universal agent for work," saying you can hand it answering questions, knowledge work such as research and document creation, and coding, all from a single input box.

The announcement

Gemini agent understands the context of a company's work, plans out the steps of a task, and gets the job done using the tools and internal systems it needs. It delivers finished work back where you already work: in your documents, your email inbox or your development environment.

Because the work keeps running in the cloud, jobs that take hours or even days continue after you close your laptop. According to Google, it has the following four types of memory and learns what the work involves and how to do it.

Memory of the task it is currently working on (retained even for tasks that run for days)
Knowledge it builds up by reading documents and interacting with people and other agents
Memory of how to do the work (including "skills" it creates itself)
A record of work it has done in the past

Three ways of working

Gemini agent works in three ways, depending on how it is used.

Personal assistant: With awareness of your calendar, team and projects, it helps with things like scheduling meetings and drafting documents.
Coworker agent: A team member with a defined role, such as project management or analysis, that works on an ongoing basis. It has its own dedicated email address, such as "@agents.company.com", and its own storage, and can access only the information that you or your team give it.
Subagents: When given a complex job, Gemini agent creates multiple temporary agents dedicated to that job, runs the work in parallel or in sequence, and combines the results.

Where it works and which AI models it uses

Gemini agent can be used directly inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. It is also available on the web, in iOS and Android apps, in Windows and Mac desktop apps, from the command line, and from Microsoft 365 and Slack, and through the API it can run headlessly inside other apps.

For AI models, it switches between not only Google's Gemini but also Anthropic's Claude. It automatically chooses the model with the best balance of performance and cost for each task, and Google plans to add support for models from other companies and open models in the future. You can also set a cap on AI spending for each project so it is not exceeded.

Security and administration

Administrative features for enterprise use are also provided.

Each agent is assigned a unique ID that prevents impersonation, and is managed in the same way as an employee
All traffic going into and out of agents passes through Agent Gateway, which applies company policies
Work is carried out in a dedicated environment with restricted external communication (Agent Sandbox)
Every agent action is logged, with a record of which agent performed it
Each agent's permissions are configured in fine detail after approval by the company's security administrators

Agents for specific industries

Google also announced previews of Gemini for the financial and legal industries.

Financial services: Comes with more than 50 built-in skills and can use data from FactSet, LSEG and S&P Global as well as US corporate filings.
Legal: Carries over the access permissions and information barrier rules of document management services such as NetDocuments and iManage as they are.

Google says versions for government, healthcare and retail will also be available soon.

Availability

Gemini agent is currently in a limited preview (private preview) for select companies. Google plans to begin a broad rollout soon for customers on some Google Workspace Business and Enterprise plans. Pricing details and availability in specific countries and regions, such as Japan, have not been announced.

CoAI's take

The competition in enterprise AI is shifting from asking AI questions and getting answers to assigning it a role, handing it work and letting it carry that work out over several days. "Coworker" agents with their own email addresses and storage are a clear sign of that shift. At the same time, because AI will handle a company's internal email and documents and interact with other people, deciding what information and permissions to give it becomes a major issue in adoption. That is likely why Google is putting security and administration features front and center. Google has not said this time whether it will be available in the consumer Gemini app.