Google integrates the use of the computer in Gemini 3.5 Flash to create autonomous agents that interact between platforms

Google has updated its model Gemini 3.5 Flash with computer use function natively integrated, so developers can create custom agents that can see, reason, and act across browser, mobile, and desktop environments.

The technology company launched Gemini 3.5 Flash last May within the framework of its annual Google I/O developer conference, a model that combines frontier intelligence with action, thus standing out in programming, agent capabilities in the real world and use of tools.

Now, Google has expressed its intention to go further and offer “the best performance” for tasks. computer use with agents, upgrading Gemini 3.5 Flash with this capability built in nativelyas shared in a statement.

The computer use function is an advanced capability that allows artificial intelligence (AI) control the web browser autonomously. Thus, it is based on the use of autonomous agents, which they see the user interface and interact with it in a real wayscrolling or typing text.

With this in mind, with the desktop tool built into 3.5 Flash, developers will be able to use Gemini’s more advanced capabilities to generate custom agents and automate tasks. These agents may see, reason and carry out actions completely autonomously both in browsers, such as in mobile and desktop environments.

This capability was only available as a stand-alone Gemini 2.5 computer usage model, however, it is now integrated directly into the main Gemini Flash model, as the technology company has highlighted.

Therefore, Google ensures that it provides a Improved performance when carrying out business automation tasks and long-term taskssuch as continuous software testing and knowledge work in professional applications.

As he has exemplified, the use of the computer is useful to generate autonomous agents that analyze what is displayed in an app running on a smartphone and return a categorized list of functionsor perform documentation audits to detect accessibility problems.

MEASURES TO ENSURE SECURITY WITH FLASH 3.5

The use of agents capable of interacting with real browser and mobile environments can also cause some cybersecurity risksas is the case of malicious code injection. Faced with this type of actions, Google has stressed its intention to offer a safe experience for users.

In this sense, to mitigate possible injection attempts, the company has indicated that they have used “targeted adversary training” for computers using Gemini 3.5 Flash. In addition, it has also launched two protection systems aimed specifically at companies.

First of all, an option has been enabled that requires explicit confirmation of the user so that the agent can carry out delicate or irreversible actions. The second option is a configuration that automatically stops the task if a direct warning injection is detected.

Likewise, Google has recommended that developers adopt a “defense in depth” approach, combining the security features mentioned with secure sandboxes, human verification and strict access controls.

All in all, the new computer usage feature in Gemini 3.5 Flash is now available to developers through the API by Gemini and the platform Gemini Enterprise Agent.