Google has integrated computer use as a built-in capability in Gemini 3.5 Flash, marking a shift from its previous standalone availability in Gemini 2.5. The update enables developers to create agents that can visually perceive, reason, and execute actions across browser, mobile, and desktop platforms through standard API access.

The integration addresses long-standing limitations in agentic workflows by combining computer use with existing function calling strengths. Enterprise users gain improved performance for automation tasks like continuous software testing and knowledge work across professional applications, with safety measures including adversarial training and optional confirmation prompts for sensitive actions.

Two enterprise safeguard systems provide defense-in-depth protection: explicit user confirmation for irreversible operations and automatic task termination when indirect prompt injection is detected. Google recommends combining these with sandboxing and strict access controls for production deployments.

Early adopters are already leveraging the capability for tasks like app feature categorization and documentation accessibility auditing. The feature launches through the Gemini API and Enterprise Agent Platform, with demo access provided via Browserbase integration.