Gemini Agent: Google Cloud launches a universal task agent that can handle tasks spanning several days and has its own Gmail account

Unveiled at Gemini at Work 2026, the agent is given objectives rather than instructions, coordinates sub-agents, chooses between Gemini and Claude, and operates within Workspace, Microsoft 365 and Slack.

At its Gemini at Work 2026 event, Google Cloud unveiled the Gemini agent, presented as a single, universal agent for enterprise work. The idea: to give it a goal rather than a series of instructions. It answers questions, takes over office work, creates media, writes and executes code, and can act for a person, for a team (like a project manager), or for a given role, like a finance analyst.

Multi-day jobs, on the most suitable model

The agent divides tasks into sub-agents that he coordinates, in parallel or in a row, over periods of time that can reach several hours or several days. Each task is assigned to the model considered most suitable: the Gemini and Claude families today, other proprietary or open models later, with a stated objective of quality and reduced costs. It runs in the cloud with a unique memory and a unique personalization graph, and each agent receives his or her own Workspace account — Gmail address, calendar and Drive space.

It is invoked by a @Gemini mention in Gmail, Docs, Sheets, Slides or Chat, but also on the command line, from Slack or Microsoft 365, and an API allows it to be integrated without a dedicated interface. The product is in private preview; expanded availability is announced soon for certain Workspace Business and Enterprise offers.