Introducing computer use in Gemini 3.5 Flash

For AI practitioners, this transition from standalone experimental models to a native, high-performance tool represents a major maturation in agent technology. By enabling models to see, reason, and interact with professional applications, developers can now automate long-horizon workflows that were previously unreachable. This capability is set to redefine enterprise automation, shifting the focus from static function calling to dynamic, cross-platform task execution that mimics human computer interaction.

The core breakthrough is the native integration of computer use into the Gemini 3.5 Flash model, replacing the previous reliance on a standalone, separate model. This update provides developers with a more efficient and powerful way to build agents capable of performing continuous software testing and complex knowledge work across diverse digital environments. Furthermore, Google has introduced a robust suite of safety features, including targeted adversarial training and optional enterprise safeguards, to mitigate risks like prompt injection and unauthorized sensitive actions.

Organizations should prioritize integrating these agentic capabilities into their current automation stacks while adopting a defense-in-depth security posture. Leaders must mandate the use of human-in-the-loop verification and secure sandboxing to ensure that autonomous agents operate within strict ethical and operational boundaries. Practitioners should immediately begin experimenting with the Gemini Enterprise Agent Platform to prototype workflows that require cross-application navigation, ensuring their teams gain early experience with these powerful new capabilities.