Google adds computer-use controls to Gemini 3.5 Flash for enterprise agents

Google has made computer use a built-in capability for Gemini 3.5 Flash, allowing developers to build AI systems that can interact with software interfaces by clicking, typing, scrolling, and navigating across environments.

Gemini 3.5 Flash gets a computer-use tool

Google announced in a post on The Keyword that computer use is now built into Gemini 3.5 Flash. The company says the feature enables agents to “see, reason, and act” across browser, mobile, and desktop environments.

The Next Web reported that the capability is available through the Gemini API and the Gemini Enterprise Agent Platform, where developers can use it to build agents that operate software interfaces rather than only returning text responses.

Google’s AI for Developers documentation describes Computer Use as a tool for building control agents for browser, mobile, and desktop settings. The documentation recommends Gemini 3.5 Flash for the feature, indicating that Google is tying the capability closely to its lower-latency Flash model family.

What developers can build

The practical change is that Gemini-powered applications can be designed to perform interface-level tasks. Instead of integrating only with structured APIs, an agent can work through graphical interfaces: selecting buttons, entering text into fields, scrolling pages, and moving through multi-step workflows.

According to Google’s developer documentation, Computer Use is intended for agents that control browsers, mobile environments, and desktop environments. That makes the feature relevant to tasks where a service may not expose a complete API, or where a workflow depends on interacting with existing software screens.

The Next Web framed the release as part of Google’s effort to make enterprise customers more comfortable with agents that can take actions on their behalf. That trust question is central: once an AI system can operate a user interface, it may also be able to make consequential changes if it is not constrained.

Safety controls are part of the pitch

Google is emphasizing safeguards alongside the new control capability. In its announcement, the company says enterprise users can configure confirmations and prompt-injection protections. Google’s Gemini API documentation also lists configurable safety policies and prompt-injection detection for Computer Use.

Those controls matter because computer-use agents are exposed to the same interfaces and content that a human user might see. A malicious web page, document, or message could attempt to influence the agent’s instructions. Google’s documentation specifically refers to prompt-injection detection, reflecting a known risk for AI systems that read untrusted content while carrying out tasks.

The company’s reference to confirmation steps also suggests that developers and enterprises can require human approval before certain actions are completed. Google’s sources do not describe every implementation detail in the excerpts provided, but they position these controls as optional enterprise safeguards rather than as a replacement for careful deployment.

A wider move toward action-taking AI

Computer-use features are becoming an important part of the agent market because they let AI systems operate existing applications without waiting for every service to offer a dedicated integration. Google’s move puts that capability directly into Gemini 3.5 Flash through official tools rather than leaving developers to assemble their own screen-control layers.

For enterprises, the appeal is straightforward: software robots that can use the same interfaces employees use may be able to help with repetitive operational work. The risk is also straightforward: systems that can click, type, and submit forms need clear permissions, monitoring, and approval boundaries.

Google’s announcement and documentation show that the company is trying to address both sides. Gemini 3.5 Flash now has a computer-use path for developers building agents across browser, mobile, and desktop environments, while Google is also highlighting safety policies, confirmations, and prompt-injection detection as part of the enterprise offering.