Introducing computer use in Gemini 3.5 Flash

At a glance:

• Gemini 3.5 Flash now integrates computer use natively, replacing the standalone Gemini 2.5 model for agentic tasks.
• Developers can build cross-platform agents via the Gemini API and Enterprise Agent Platform.
• Enhanced safety measures include adversarial training and enterprise safeguards for prompt injection risks.

Integration into Gemini 3.5 Flash

The latest Gemini 3.5 Flash model embeds computer use directly into its core functionality, eliminating the need for the previously separate Gemini 2.5 computer use model. This integration allows developers to create agents that seamlessly interact across browsers, mobile devices, and desktops. The model excels at tasks requiring long-horizon reasoning, such as continuous software testing and knowledge work in professional applications. For instance, 3.5 Flash can analyze an app's documentation to generate a categorized list of features or audit its own content for accessibility issues. This native capability improves reliability and performance for enterprise automation workflows.

The shift to native integration also simplifies deployment. Developers no longer need to manage separate models for computer use tasks. Instead, they can leverage 3.5 Flash's existing strengths in function calling and tool integration (like Search and Maps) to build more cohesive agents. This reduces complexity and lowers the barrier for enterprises adopting agentic workflows.

Safety and Enterprise Safeguards

To address risks associated with autonomous agents, Google has implemented targeted adversarial training for computer use in Gemini 3.5 Flash. This training helps mitigate prompt injection vulnerabilities by teaching the model to recognize and resist malicious inputs. Enterprises can further enhance security through two optional safeguard systems: requiring explicit user confirmation for sensitive actions and automatically halting tasks if indirect prompt injection is detected.

However, Google emphasizes a "defense-in-depth" strategy. While these safeguards are robust, developers are encouraged to combine them with secure sandboxing, human-in-the-loop verification, and strict access controls. The company provides detailed best practices documentation to guide implementation. This layered approach ensures that even if one layer fails, others can prevent potential breaches. Enterprises adopting 3.5 Flash should consider these additional measures to maintain compliance and security in automated environments.

Customer Adoption and Use Cases

Several enterprises are already leveraging Gemini 3.5 Flash's computer use capabilities to drive efficiency. While specific quotes from customers aren't provided in the source, the article highlights real-world applications like automated software testing and knowledge management systems. The ability to build agents that "see, reason, and take action" across platforms opens new possibilities for industries reliant on complex workflows. For example, a financial services company might use these agents to monitor market data and execute trades based on predefined criteria without manual intervention.

The scalability of 3.5 Flash's computer use is another key advantage. Unlike previous models that required manual setup for each task, the integrated approach allows for dynamic adaptation to new environments. This makes it particularly valuable for organizations needing to automate repetitive or high-volume processes across multiple systems.

How to Start Using Computer Use Today

Developers can begin experimenting with Gemini 3.5 Flash's computer use through a demo environment hosted by Browserbase. This sandboxed platform allows users to test agent capabilities without risking production systems. For production use, integration is available via the Gemini API and Gemini Enterprise Agent Platform. The documentation includes reference implementations to help developers quickly deploy agents tailored to their specific needs.

To get started, users should first explore the Browserbase demo to understand the model's capabilities. Next, they can access the Gemini API documentation to configure their agents. The Enterprise platform offers additional tools for managing security and compliance, ensuring that large-scale deployments meet organizational requirements. Google provides ongoing support through its developer community and technical documentation.

Future Implications

The integration of computer use into Gemini 3.5 Flash represents a significant step toward more autonomous AI systems. As agents become capable of interacting across multiple platforms, their potential applications expand beyond current use cases. However, this also raises questions about accountability and control. Enterprises must balance the benefits of automation with the need for human oversight, especially in high-stakes environments.

Google's approach to safety and enterprise tools suggests a proactive stance on responsible AI development. By combining technical safeguards with best practices guidance, the company aims to make advanced agentic capabilities accessible while minimizing risks. As adoption grows, we can expect further refinements to these systems, potentially leading to more sophisticated and secure AI agents in the future.

Conclusion

Gemini 3.5 Flash's native computer use capability marks a pivotal advancement in AI agent development. By embedding this functionality directly into the model, Google simplifies deployment and enhances performance for enterprise applications. The combination of robust safety measures and practical tools positions 3.5 Flash as a powerful solution for automating complex workflows. As more developers and enterprises adopt this technology, its impact on productivity and innovation across industries is likely to be substantial.