Details
- Google for Developers announced that Gemini 3.5 Flash now includes built-in computer use capabilities, no longer requiring a separate model.
- Developers can build agents that can see UI elements, reason about them, and take actions across browser, desktop, and mobile environments using a single model.
- The update supports custom client-side functions so organizations can define domain-specific actions and allow human-in-the-loop takeover for sensitive workflows.
- Configurable, action-level safety policies let teams enforce guardrails on operations such as data entry, navigation, or file manipulation, complementing Google’s broader defense-in-depth approach.
- Computer use in Gemini 3.5 Flash is exposed through the Gemini API, Gemini Enterprise Agent Platform, and related tools, enabling automation of long-horizon tasks like software testing, knowledge work, and multi-app workflows.
- Google highlights prompt-injection protections and enterprise safeguards, including options to require explicit user confirmation for high-risk actions and to stop tasks when indirect prompt injection is detected.
- The feature builds on earlier standalone computer use models, consolidating agentic tools such as function calling, Search, Maps grounding, and UI control into the main Flash model for higher performance and simpler integration.
Impact
By folding computer use directly into Gemini 3.5 Flash, Google reduces friction for developers building agentic applications that operate real user interfaces, not just APIs. This move strengthens Gemini’s position against OpenAI and Anthropic in enterprise automation, and could accelerate adoption of AI agents for routine desktop, browser, and mobile workflows under tighter safety controls.
