Google has integrated computer use capabilities directly into Gemini 3.5 Flash, marking a significant upgrade to its flagship AI model. The feature enables custom agents to observe, reason, and act across browser, mobile, and desktop environments.
Previously, computer use was limited to a standalone Gemini 2.5 computer use model. The new integration brings these capabilities to Google's highest-performing Flash variant, available through the Gemini API and Gemini Enterprise Agent Platform.
The rollout targets developers and enterprise users seeking advanced automation solutions. Early adopters have deployed the features in live environments for tasks including continuous software testing and enterprise knowledge work.
View tweet from @osanseviero
Google has implemented two optional enterprise safeguard systems to address security concerns. The first requires explicit user confirmation for sensitive actions, while the second halts tasks if indirect prompt injections are detected.
The company recommends combining these measures with sandboxing, human-in-the-loop verification, and strict access controls. Technical upgrades include targeted adversarial training to mitigate prompt injection risks.
Enterprise automation focus
The computer use integration supports complex automation workflows that span multiple applications and environments. Use cases range from filing support tickets to conducting multi-step research tasks across different software platforms.
Google's approach emphasizes security and flexibility for enterprise customers. The safeguard systems aim to prevent unauthorized actions while maintaining the model's ability to execute sophisticated automation tasks.
The update positions Gemini 3.5 Flash to compete more directly with Anthropic's Claude, which introduced computer use capabilities in October 2024. Google's implementation focuses on multi-environment support rather than single-application control.
Google continues expanding Gemini's enterprise capabilities as organizations increasingly adopt AI-powered automation for business-critical workflows.
💬 Discussion
Sign in to join the discussion.
Sign in →No comments yet — be the first.