Google aims to change how we interact with AI. The company wants your assistant to become a real digital coworker beyond simple text boxes. With this in mind, they just rolled out a handy visual feature for everyday web browsing alongside a massive upgrade for developers. The big news here is that the Google Gemini 3.5 Flash model now has a native, built-in “computer use” tool, bringing us much closer to truly autonomous AI agents.
For everyday users, this matches the release of Chrome 149, which introduces a neat feature called “Select from screen.” You can find it right inside the browser’s attachment menu. Once you click it, it highlights your active tab and lets you drag a box over any image or text to drop it instantly into your Gemini prompt. It is a quick, effortless way to ask questions about whatever you are looking at online.
Google Gemini 3.5 Flash becomes an AI that clicks and scrolls
While the Chrome update is a great quality-of-life tweak, the developer side is where things get interesting. Google officially integrated the Computer Use tool into the main Flash model pipeline, replacing the older Gemini 2.5 testing framework. This means AI agents can now perceive and interact with software interfaces just like a human would, relying on screenshots and visual understanding rather than deep, rigid coding.
This opens many possibilities for office automation and software testing. Developers can now build custom agents that navigate websites, fill out long forms, click buttons, and handle repetitive data collection across desktop, mobile, and browser environments. To speed things up, Google opened a live demo space hosted by Browserbase, letting creators test these capabilities right away through the Gemini Enterprise Agent Platform.
Keeping the desktop safe
Giving an AI control over a mouse and keyboard naturally brings up safety questions. If an agent lands on a sketchy website with hidden text commands, it could trigger actions you did not intend. Google addressed this directly, noting that they used targeted adversarial training to protect the model against indirect prompt injections.
On top of that, they launched two enterprise safety systems. Companies can set up the software to require explicit human approval before the AI makes any sensitive or permanent changes. The system can also automatically freeze active tasks the moment it spots an incoming prompt attack. This way, Google aims to deliver a tool that handles real work without putting your desktop at risk.
The post Gemini 3.5 Flash Can Now Control Your Desktop to Handle Boring Software Tasks appeared first on Android Headlines.
