Google launches a task-planning agent in Gemini, aimed first at businesses

Google is adding agent-style features to Gemini that let the AI plan and carry out multistep work on its own, instead of only answering questions. The company announced the change at a Google Cloud event on Thursday, and said it will roll the capability out first to business customers before offering it to consumers.

Google Cloud CEO Thomas Kurian said the new agent can be given “objectives, not just instructions.” In practice, that means a worker can describe a goal, and the agent plans the steps, pulls in custom skills and tools, and connects to a company’s internal systems to get the work done. Users can attach files, folders, or whole projects to a request. By default, Gemini picks the model it thinks best fits the task, but people can override that choice, including picking a model from a third party. Google said Anthropic’s Claude models are the first outside option, with open-source and other private models to follow. A “tasks inbox” lets people watch the agent’s reasoning, see it hand off work to subagents, and track its progress.

An employee with its own inbox

The most unusual part of the launch is identity. Google said the agent gets its own Google Workspace account, complete with an email address and its own context about the company: who sits on which team, what time zone they work in, who needs to sign off on what, and what is on people’s calendars. Employees can reach it by tagging it, emailing it, sharing a document with it, or adding it to a group chat. When it takes an action, Google records an audit trail attributed to the agent rather than to a person.

The agent can connect to Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, and Snowflake, along with any Model Context Protocol (MCP) server, a standard for linking AI tools to outside data, whether it sits inside or outside the company’s network. Google is making it available on iOS and Android, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365, ServiceNow, and Slack.

Sundar Pichai framed the enterprise-first rollout as a way to work through “harder problems around security, scale, and performance” before a wider launch. He pointed to Gemini’s existing reach as a reason the company feels ready to push further into agentic AI: more than 1 billion monthly active users, and adoption by nearly 90% of Fortune 100 companies through Gemini Enterprise. Google named On, Shopify, and PayPal as early testers, and listed BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers among its Gemini Enterprise customers.

Google also announced new spending controls for businesses worried about agentic AI costs, including multi-model orchestration, smart routing between models, and real-time spend caps. The company has not said when the agent will reach individual consumers, or detailed how the security and performance questions Pichai raised will be resolved before that happens, and it has not specified pricing for the new capabilities within Gemini Enterprise. Whether competing agent products from other companies change that timeline is something to watch as Google expands testing with its named enterprise customers.