Google Cloud unveiled the Gemini agent on Thursday, a universal AI tool designed to execute workplace tasks autonomously. Presented by CEO Thomas Kurian at the Gemini at Work event, the agent operates within Gmail, Docs, Sheets, Slides, Drive, Chat, and Calendar, while also integrating with Microsoft 365 and Slack. Unlike traditional assistants, this agent runs in the cloud for hours or days after a user closes their laptop, allowing it to complete complex goals independently.
A key feature is the creation of "coworker agents," which managers can assign specific roles such as events coordinator. These agents receive their own Workspace account, including a unique email address, calendar, Drive storage, and a listing in the company directory. To ensure security and accountability, each agent is assigned a cryptographically attested identity, and all actions are recorded in an audit trail distinct from human users. The system utilizes an Agent Sandbox for containment and an Agent Gateway described as an "AI network firewall." Google noted that nearly 500 customers processed more than one trillion tokens over the past year, prompting the addition of real-time spend caps. Pricing and general availability dates were not disclosed, though financial services and legal versions are in preview.
The introduction of agents with dedicated email addresses and directory listings marks a structural shift in how enterprises manage digital labor. By assigning cryptographic identities and separate audit trails, Google attempts to integrate autonomous software into existing compliance frameworks that traditionally rely on human attribution. This approach addresses immediate regulatory concerns regarding accountability, ensuring that actions taken by AI are traceable and distinct from employee activities. However, the reliance on sandboxing and gateway firewalls highlights ongoing operational risks, particularly given recent reports of containment failures in similar agentic systems.
Market implications center on the transition from token-based billing to outcome-based automation, evidenced by the implementation of real-time spend caps. As agents consume significant computational resources, cost control becomes a primary governance challenge for IT departments. The lack of published pricing suggests Google is still calibrating the economic model for sustained autonomous operations. Furthermore, the orchestration across multiple models, including Anthropic’s Claude, indicates a move toward vendor-agnostic infrastructure, potentially reducing lock-in effects but increasing complexity in quality assurance and security oversight.
