Google has moved one of the Gemini API’s agent tools to general availability.
The news. According to the Gemini API release notes, Google released the Computer Use tool to general availability on October 30, 2026. The entry says GA covers both the Interactions API and the GenerateContent API.
What the tool does. Per Google’s Computer Use guide, the tool lets developers build agents that control a browser, a mobile device or a desktop. The model “sees” the screen through screenshots and “acts” by producing specific UI actions, such as mouse clicks and keyboard input.
Who actually clicks. The model does not touch your computer. The guide compares the setup to function calling: you build the client-side environment that receives the proposed actions and carries them out. Each step is a loop. Your app sends the prompt, the tool configuration and a screenshot.
The model returns a suggested action, such as a click, scroll or keystroke. Your code carries it out, captures a fresh screenshot and sends that back, and the cycle repeats until the task ends.
Newer model features. The guide says Gemini 4 and Gemini 3.x models support several extras: agents for browser, mobile and desktop environments; an intent field that explains the reasoning behind each action; configurable safety policies with built-in categories and overrides; and opt-in screenshot scanning to detect hidden prompt injection instructions.
Google’s own caution. GA does not mean hands off. The guide warns that Computer Use may contain errors and security vulnerabilities. Google recommends close supervision for important tasks and advises against using it for critical decisions, sensitive data, or actions where serious mistakes cannot be undone.
Suggested uses. The guide lists repetitive data entry and form filling, automated testing of web apps and user flows, and research across websites, such as gathering product prices and reviews.
What’s real here: the tool is now a supported part of the Gemini API, not an experiment. But the developer’s app still executes every action, so the safety and supervision work stays with whoever builds the agent.
Sources
By Willa Ravenwire · The AI Bandwagon
