Google Cloud has announced “Gemini agent,” a general-purpose agent for businesses. It offers work-focused and governance features, including the ability to use Gemini and Claude selectively.
Google Cloud announced “Gemini agent,” a general-purpose agent for businesses, on 8 October. It said a single agent can handle question answering, knowledge work, image and media generation, and code creation and execution, and it was reported to currently use Google’s Gemini models and Anthropic’s Claude selectively.
The company positions it as “a single general-purpose agent for work,” explaining that chat, automated tasks and code generation can all be accessed through the same interface. It can be called from the web, iOS, Android, Windows and Mac environments, the command line, Google Workspace, Microsoft 365 and Slack, and can also be embedded in other applications in headless mode. Processing continues in the cloud, and some tasks can continue running for several hours to several days after the device is closed, the company said.
The company described a “Coworker Agent” multi-agent setup that dynamically creates sub-agents according to the task. It listed Confluence, Teams, Slack, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake as integrations, and explained that the system can also connect to any MCP server. It runs directly across Google Workspace’s Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar as well.
For governance, it provides identity and access management, audit logs and a sandbox for each agent. It also offers Agent Gateway, which functions as an AI network firewall. For cost control, users can set automatic model routing through Smart Routing and real-time spending caps at the project level, with agents stopping when a cap is reached.