Google has officially changed the AI agent race. In a major upgrade launched today, native “computer use” capabilities are now live within the flagship Gemini 3.5 Flash model.
Announced by Mateo Quiros, Product Manager at Google DeepMind, this feature transitions screen-interaction directly into Google’s fastest model. Instead of relying on clunky, separate systems, developers can build custom agents that see, reason, and interact with operating systems natively. It marks a huge step forward in the Google AI News hub, pushing the boundaries of what autonomous mobile and desktop software can achieve.
The Core Concept: You give the agent a screen and a goal—it figures out the actions.
One Model, Three Environments: Web, Mobile, & Desktop
By integrating computer use directly into Gemini 3.5 Flash, Google delivers Pro-level execution and blazing speed at a fraction of the cost of standalone model setups. As shown in the benchmark chart within computer use in Gemini 3.5 Flash.jpg, Gemini 3.5 Flash achieves an impressive 78.4 score on OSWorld-Verified tasks, matching premium models like Sonnet 4.6 while drastically outperforming Gemini 3 Flash (65.1).
This upgrade transforms Gemini 3.5 Flash into a cross-platform powerhouse, supporting three major environments:
Web Browsers: Navigating complex web apps, clicking buttons, and managing forms.
Mobile (Smartphones): Interacting with mobile operating systems and simulating touch inputs.
Desktop Environments: Controlling desktop software, moving mouse pointers, and typing based on real-time screenshots.
How the “Agentic Loop” Works
Screenshot: The client application captures the current screen GUI.
Analyze: Gemini 3.5 Flash reads the pixels and plans its next step.
Action: The model outputs a precise UI command (like clicking exact X/Y coordinates).
Repeat: The environment executes the command, snaps a new screenshot, and repeats until the goal is met.
Why This Matters
Most business software and online services were designed for human interaction rather than AI. Computer Use bridges that gap by allowing Gemini agents to work directly within existing interfaces.
Instead of building dedicated API integrations for every application, developers can deploy AI agents that visually understand software and complete tasks using the same controls available to human users.
Google says Gemini 3.5 Flash is specifically optimized for long-horizon agentic tasks, making it suitable for workflows that require planning, decision-making, and multiple sequential actions.
Enterprise and Developer Benefits
For enterprises, Computer Use could dramatically reduce manual work across departments.
Potential use cases include:
Customer Support Automation
AI agents can navigate support dashboards, retrieve information, update records, and assist customers across multiple systems.
Data Entry and Back-Office Operations
Organizations can automate repetitive administrative tasks without redesigning existing software infrastructure.
Software Testing
Developers can create AI-powered testing agents that interact with applications similarly to real users.
Research and Information Gathering
Agents can browse websites, collect data, summarize findings, and prepare reports with minimal supervision.
These capabilities align with Google’s broader push toward agent-based AI platforms announced alongside Gemini 3.5 Flash.
Gemini 3.5 Flash: Built for the Agent Era
Google introduced Gemini 3.5 Flash as the first model in the Gemini 3.5 family, focusing heavily on coding, reasoning, and autonomous task execution.
According to Google, the model delivers frontier-level intelligence while maintaining the speed and efficiency required for real-world agent deployments. It excels in:
- Complex coding workflows
- Multi-step reasoning
- Tool usage
- Long-context understanding
- Autonomous task planning
The model supports large context windows and is designed for rapid agentic loops where AI systems continuously plan, execute, evaluate, and improve their actions.
Competition in the AI Agent Race
Google’s Computer Use announcement arrives amid intense competition among AI companies racing to create fully autonomous digital agents.
The industry is increasingly shifting from conversational AI toward systems capable of performing meaningful work on behalf of users. Computer Use represents a key building block for that future, allowing AI to interact with software environments that were never designed for machine automation.
With Gemini 3.5 Flash, Google is positioning itself as a major contender in the emerging AI agent ecosystem.
Availability
Google is making Gemini 3.5 Flash available across its AI ecosystem, including developer platforms and enterprise offerings. Developers can leverage the model through Google’s AI tools to build next-generation autonomous agents capable of understanding, reasoning, and acting across digital environments.
The Bigger Picture
Computer Use may ultimately become one of the most important AI advancements of 2026. By enabling Gemini to directly operate software and websites, Google is moving AI beyond assistance and into execution.
As businesses increasingly seek automation without costly software rewrites, AI agents capable of using existing applications could unlock a new wave of productivity gains. Gemini 3.5 Flash’s Computer Use capabilities signal that the era of truly autonomous AI workers is rapidly approaching.
How to Get Started Today
The native computer use tool parameter is available starting today via the Gemini API and the Gemini Enterprise Agent Platform.
Test the Demo: Prototyping workflows can be done immediately in a ready-to-use sandbox hosted by Browserbase.
Start Building: Developers can access Google’s reference implementation today to begin enabling the tool in their standard
gemini-3.5-flashAPI calls.
For more deep-dives into the ecosystem, check out our full coverage of Google’s latest AI updates and releases.
