**[Google's Flagship Model Gemini 3.5 Flash Natively Supports Computer Control, Unlocking Enterprise-Grade Agent Automation]** According to monitoring by Beating Insights, Google has integrated the computer control (Computer Use) feature as a built-in tool directly into its flagship large model Gemini 3.5 Flash. Prior to this native integration, developers had to invoke the specialized Gemini 2.5 computer control model to execute agent tasks. With the native integration, developers and enterprise users can now directly use the Gemini API or the Google Cloud Gemini Enterprise Agent Platform (formerly known as the Vertex AI platform) to enable the flagship model to control devices, simplifying the agent development architecture. The built-in computer control tool processes screenshots from browser, mobile, or desktop environments to perform visual perception and step reasoning, subsequently outputting operational commands such as mouse clicks, keyboard inputs, scroll wheel movements, and menu navigation. These commands facilitate long-process automation tasks like continuous software testing and cross-webpage data collection. To aid debugging and auditing, the model includes an "intent" field when generating commands to explain the logic behind each step. Addressing potential prompt injection risks that agents may encounter in real-world network environments, Google has conducted targeted adversarial training on the model and offers two optional safeguards: mandatory human approval for irreversible operations such as financial transactions or file deletions, and automatic task termination if indirect injection commands are detected in screenshots. Currently, Browserbase provides an online hosted demo environment (gemini.browserbase.com), and Google has simultaneously open-sourced a reference implementation code named "computer-use-preview" on GitHub. [Original Link]