Google wants the chat window to stop being the place where AI work ends. On Thursday, Google Cloud CEO Thomas Kurian introduced the Gemini agent at the Gemini at Work 2026 event. It is a single agent meant to take a goal, work on it in the cloud and hand back a finished result.
What the Gemini agent does
Google describes it as a universal agent for work. Users state an objective instead of issuing step-by-step commands, and the agent handles research, documents, image and media generation, and writing and running code. Because it runs in Google’s cloud, tasks can continue for hours or days, even after the user closes a laptop.
It can be called from Gmail, Docs, Sheets, Slides and Chat by mentioning @Gemini, and it is also reachable from the command line, Slack and Microsoft 365. An API lets other companies embed it as a “headless” agent with no dedicated interface.
Memory and sub-agents
The agent keeps one set of memories and one personalization graph across devices. Google splits memory into four types:
- session memory for the current task
- semantic memory for knowledge built from documents and conversations
- procedural memory for how jobs get done, including skills the agent writes for itself
- episodic memory for everything it has done before
For bigger jobs, Gemini can create temporary sub-agents, each with its own identity, and coordinate them in parallel or in sequence. Teams can also set up persistent “coworker” agents with defined roles, their own company email addresses, calendar and storage. Google says these agents only see the context they are given.
Gemini and Claude under one roof
The agent routes each job to a model that fits it. Simple tasks may go to Gemini Flash, while long-horizon work can run on the flagship Argon. Anthropic’s Claude models are already available, and Google says other closed and open models will follow.
Enterprise controls come with it: agent identity, role-based permissions, activity logging and cost-control tools.
A crowded race
Google is not alone here. VentureBeat’s comparison places the Gemini agent next to OpenAI’s Dots and ChatGPT Space and SpaceXAI’s Grok Bot, which are also pitched as persistent, always-on coworkers. SiliconANGLE calls persistent execution a common trend across the industry.
Based on the launch coverage, Google’s differentiator is the combination of deep Workspace and Microsoft 365 integration with openness to a rival’s models. Figures cited at the event say nearly 80% of Google Cloud customers already use its AI products and nearly 90% of the Fortune 100 use Gemini Enterprise, so the distribution is already in place.
The open question is oversight. Agents with their own mailboxes and days-long autonomy need tight limits, and Google’s pitch leans heavily on permissions and logging. How well those controls hold up in real companies will matter more than the demo.