The Google Cloud division presented a new product at the Gemini at Work 2026 conference called Gemini agent – an artificial intelligence agent designed for enterprise users. It is available in the Gemini Enterprise application, Workspace package and a number of other services. He has enough “set goals, not give step-by-step instructions” — Gemini agent can answer questions, perform intellectual work, create media content, write and run code.
Image source: Google
“If your manager sends you an email asking you to prepare a presentation with the latest information on a project, Workspace Intelligence identifies it as a task that can be delegated and offers you the click of a button to transfer it to a Gemini agent.”— gives an example Google. The AI agent can also organize an event with colleagues by independently determining their composition from previous events, checking their calendars and initiating correspondence to agree on a convenient time.
Gemini agent is able to act as an employee of the company and perform the functions provided for a specific position. It performs tasks by selecting the most suitable models – in addition to the Google Gemini family, these are also Anthropic Claude models; in the future there will be others, proprietary and open. To solve multi-stage problems, subagents are connected; Supports work with various third-party tools. Tasks can be long-term and take several hours or even days.
The AI agent runs in the cloud, accessing “a single set of memory, context, and overall personalization graph.” Four memory types are available:
- session memory – focused on the current task, even if it has been running for several days;
- semantic memory – “a structured knowledge base formed in the process of reading documents, communicating with people and interacting with other agents”;
- procedural memory – “information about how work is done, including the skills that the agent develops for himself”;
- episodic memory – “everything that has been done before”.
To invoke the Gemini agent, it must be mentioned as “@Gemini” in Gmail, Google Docs, Sheets, Slides, and Chat. It is available on any device through various channels, including the command line, Google Workspace, Slack and Microsoft 365; Using the API, you can connect third-party applications, including those without a graphical interface. Currently, Gemini agent is at the stage of closed preliminary testing; in the future, it will be available to corporate clients of Google Workspace on some tariff plans of the Business and Enterprise families.
If you notice an error, select it with the mouse and press CTRL+ENTER.
