Google announced a unified agent inside Gemini that can plan tasks and execute them across enterprise applications and systems, with the ability to delegate work to sub-agents and use models from Google and other providers. The rollout will begin with enterprise customers before the technology is offered to consumers, with an independent enterprise identity and an audit log of actions.
Google announced during a Google Cloud event that it is transforming Gemini from a tool for answering questions into an agent capable of taking on complete goals and executing them on the user’s behalf. The agent can plan the work, use custom skills and tools, and connect to the company’s internal systems through a unified interface.
Google will focus initially on enterprises before making the agent available to consumers. The company’s CEO, Sundar Pichai, said that more than one billion monthly active users use Gemini, and that nearly 90% of Fortune 100 companies use Gemini Enterprise at work. According to Google, focusing on businesses makes it possible to address the security, scale, and performance challenges associated with operating agents capable of taking actions.
From Instructions to Goals
Thomas Kurian, CEO of Google Cloud, explained that the agent can be given goals rather than only separate instructions. Gemini then plans the required steps, uses the available skills and tools, and connects the work to the enterprise’s systems to achieve the goal.
Gemini selects the most suitable model for the task by default, but users can choose another model, starting with Anthropic’s Claude models. Google plans to add open-source models and other proprietary models to the model selector in the future. Users can also attach files, folders, and custom projects to specific workflows.
Broad Connections and an Independent Identity
The agent can connect to services such as Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, and Snowflake, in addition to Model Context Protocol (MCP) servers inside or outside the company’s network, according to Google.
The agent will receive an independent Workspace account, including an email address and its own context. In this capacity, it can be called by mentioning it, contacted by email, or included in a group conversation. When executing actions, it also creates an audit log attributed to the agent rather than to a human employee.
Users will be able to monitor what Gemini is executing through a “task inbox” interface that displays work progress, delegates tasks to sub-agents, and loads custom skills, the code used, and the execution path.
What Changes in Practice?
The announcement moves Gemini from an assistant model that waits for a direct request to a system capable of coordinating multiple steps and working across different sources and systems. This means the tool’s value will depend not only on the quality of the model, but also on the precision of permissions, the limits on access to enterprise data, and the ability to review the actions executed by the agent.
The agent will be available on iOS and Android devices, Windows and Mac devices, the command-line interface, Google Workspace, Microsoft 365, ServiceNow, and Slack. Early testers include sportswear company On, Shopify, and PayPal, while Google said that Gemini Enterprise customers include BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers.
Google also announced flexible spending options that include coordinating multiple models, smart routing, and immediate spending caps, with the aim of helping companies control enterprise AI costs. The source did not provide details about pricing, the general availability date, or specific permission limits—points that institutions will need to review before adopting the agent broadly.