Google adds screen-control tool to Gemini 3.5 Flash

Google launched computer use as a built-in tool in Gemini 3.5 Flash, letting AI agents interact with screens natively.thenextweb+1
The capability previously required a separate model; it now works alongside code execution, search, and function calling in a single model.thenextweb+1
Google added enterprise safeguards including user confirmation for sensitive actions and automatic halts against prompt injection attacks.thenextweb+1

Google Alphabet Inc. on Tuesday launched computer use as a built-in tool in Gemini 3.5 Flash, enabling developers to build AI agents that can see, reason about, and interact with screens across browser, mobile, and desktop environments. The capability, previously available only through the standalone Gemini 2.5 computer use model, is now natively embedded in the main Flash model.thenextweb+1

From Standalone Model to Native Tool

The integration means developers no longer need to call a separate, dedicated model to build agents that interact with graphical interfaces. Instead, computer use can be activated as one of several tools within Flash — alongside code execution, search, and function calling — through the Gemini API and the Gemini Enterprise Agent Platform, formerly known as Vertex AI.investing+1

Product manager Mateo Quiros described the integration as giving Flash the ability to "see, reason about, and take action on screens," according to The Next Web. Google said the update improves performance for long-horizon and enterprise automation tasks such as continuous software testing and knowledge-based work within professional applications.thenextweb+2

Enterprise Safeguards and Security

To address security risks for agents operating in live environments, Google introduced focused adversarial training to counter prompt injection attacks. The company is also offering two optional enterprise safeguards: one that requires explicit user confirmation before the agent executes sensitive or irreversible actions, and another that automatically halts the agent if it detects an indirect prompt injection attempt.investing+1

Google is additionally providing a demonstration environment hosted by Browserbase for developers to test these capabilities.investing

Context Within Google's AI Strategy

Gemini 3.5 Flash was first released at Google I/O on May 19, 2026 as Google's fastest agentic AI model, delivering what the company called frontier performance for agents and coding tasks. The model launched with a one-million-token input context window and native support for function calling, structured output, and code execution, but did not initially include computer use.simonwillison+3

The June 24 update, documented in Google's API changelog as a public preview, fills that gap and positions Gemini 3.5 Flash as a single model capable of handling the full range of agentic tasks — from code generation to direct screen interaction — without requiring developers to orchestrate multiple models.google

Sources (14)