Google Cloud has unveiled Gemini Agent, a universal AI tool designed to handle workplace tasks across applications like Gmail, Docs, Sheets, and external platforms including Microsoft 365 and Slack. The system operates in the cloud, allowing tasks to continue even after a user disconnects. CEO Thomas Kurian introduced the product at the Gemini at Work event.
A key feature allows users to create 'coworker agents' by defining a specific role. Each agent receives its own Workspace account, including a unique email address, calendar, and directory listing. These agents integrate with third-party services such as Salesforce, Jira, and Snowflake via the Model Context Protocol, and can utilize both Gemini and Anthropic Claude models.
To address security, Google assigns each agent a cryptographically attested identity and maintains an audit trail for all actions. The agents run within an 'Agent Sandbox' and communicate through an 'Agent Gateway' firewall. These measures aim to isolate the software from the broader system, following previous industry reports of AI agents escaping virtual environments.
Google has implemented real-time spending caps to manage costs associated with token consumption. The company noted that nearly 500 customers processed over one trillion tokens in the past year. If a project reaches its predefined spending limit, the agent automatically pauses operations until a user manually resumes the task.
Currently, versions of the agent tailored for financial services and legal sectors are available in preview. Google plans to expand support to government, healthcare, and retail industries in the near future. The company has not yet disclosed official pricing or a date for general availability.
