Google Cloud has introduced a new Gemini agent designed to take on a wide range of workplace tasks, from answering questions and creating presentations to writing code and coordinating work across business applications.
Announced at the company’s Gemini at Work 2026 event, the agent represents Google’s effort to move Gemini beyond a chatbot that responds to prompts and toward an AI system that can be given an objective and carry out the steps needed to complete it, according to a company blog post.
“You give it objectives, not instructions,” Google Cloud CEO Thomas Kurian wrote. “You delegate an outcome and come back to finished work.”
In practical terms, an employee could ask Gemini to research market trends, build a financial model in Google Sheets and then turn the findings into a presentation without having to separately instruct the AI at every stage.
Gemini can also work directly inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. Google said the agent carries its memory and context across those applications, allowing it to understand relationships among a user’s projects, colleagues and documents.
The agent can also keep working after a user closes a computer. Because tasks run in the cloud, Google said jobs can continue for hours or days. For more complicated assignments, Gemini can create temporary sub-agents to handle different pieces of the work in parallel or in sequence.
Google is taking the concept a step further with “coworker agents,” persistent AI workers assigned specific roles. A company could create an agent for a team that receives its own email address, calendar, Drive storage and entry in the company directory. Employees could then email or @mention the agent much as they would another colleague.
Despite the Gemini name, the agent is not limited to Google models. Google said it can choose among its Gemini models and Anthropic’s Claude models depending on the task, with additional private and open models planned. The goal is to use more powerful models for difficult work while routing simpler tasks to less expensive models.
Google also introduced real-time spending caps that allow companies to place hard limits on AI costs for individual projects. If a project reaches its limit, the agent pauses until the company chooses to resume it.
The launch puts Google more directly into the intensifying competition to build AI agents capable of carrying out work across applications rather than simply generating answers. The announcement follows recent agent launches from OpenAI and Meta.