Google integrates computer use into Gemini 3.5 Flash, enabling AI agents to control screens

Google Integrates Native Computer Use into Gemini 3.5 Flash

Google has announced that computer use capability is now a built-in tool in Gemini 3.5 Flash, marking a shift from requiring a dedicated standalone model to having the functionality integrated directly into its fastest agentic model1

1

. The feature enables developers to build AI agents that can see, reason, and take action across browser, mobile, and desktop environments without calling a separate computer use model2

2

. Previously, this capability was only available through a standalone Gemini 2.5 computer use model that Google first released in October 2025, which achieved roughly 70 percent accuracy on the Online-Mind2Web benchmark2

2

.

The integration means developers can now activate native computer use as one of several tools within Gemini 3.5 Flash, alongside code execution, search, and function calling2

2

. According to Mateo Quiros, Product Manager at Google DeepMind, this delivers Google's best performance yet for agentic computer use tasks3

3

. The feature is currently available to developers and enterprise customers via the Gemini API and the Gemini Enterprise Agent Platform1

1

.

AI Agents Can Now Click Buttons and Fill Forms Autonomously

This update positions Gemini beyond a simple chat and text-generation platform. AI agents can now click buttons, fill forms, and complete complex tasks by understanding what is happening on a screen and responding to it4

4

. The AI agents built on Gemini 3.5 Flash can autonomously navigate and perform tasks such as finding flight deals, playing games, or extracting data from dashboards1

1

. To demonstrate its capabilities, Google created a Browserbase instance where users can prompt the model to perform tasks—one example showed the model visiting three separate flight booking websites, entering dates, searching through tickets, and returning the best options1

1

.

The ability to see and control your screen opens possibilities for enterprise automation tasks including continuous software testing, where agents navigate applications and verify functionality without human testers stepping through each screen2

2

. Knowledge workers could deploy agents to complete multi-step browser tasks, fill forms, perform data extraction from dashboards, or navigate internal tools2

2

. For businesses, this can save time and reduce manual effort on repetitive office work, scheduling, and data entry4

4

.

Enterprise Safeguards Address Prompt Injection Risks

The ability for AI agents to control computers and perform tasks autonomously raises questions around safety, especially for enterprise consumers. Google has implemented targeted adversarial training specifically for prompt injection—the attack where malicious instructions embedded in a webpage or document trick an AI agent into performing unintended actions2

2

5

5

. To mitigate these risks, Google is introducing two optional enterprise safeguards built into computer use with Gemini 3.5 Flash1

1

.

The first safeguard requires explicit user confirmation before the agent executes any action flagged as sensitive or irreversible, such as submitting a form, making a purchase, or deleting data2

2

. The second automatically halts the agent if it detects an indirect prompt injection attempt, stopping execution rather than risking a compromised action2

2

. Both safeguards are opt-in rather than defaults, and Google recommends a defense-in-depth approach where developers layer multiple protections2

2

. The company specifically advises combining these safeguards with secure sandboxing, human-in-the-loop verification, and strict access controls3

3

.

Competitive Landscape and Enterprise Adoption Questions

The competitive landscape has shifted considerably since Anthropic pioneered the category with Claude Computer Use, which works across operating systems and can interact with file systems beyond just browsers2

2

. OpenAI has also entered the space, and the three companies are now competing on different dimensions. The question for enterprise buyers centers less on which model can click a button and more on which one can do it safely inside regulated environments2

2

.

Google has not published updated benchmark scores for computer use as a built-in Flash tool versus the previous standalone model, nor disclosed how many enterprises are using the capability or provided case studies with named customers2

2

. The Gemini Enterprise Agent Platform uses pay-as-you-go pricing, and Flash is one of the cheaper models in Google's lineup, which could make computer use more accessible for large-scale automation2

2

. Developers and enterprises can start using the feature through a Browserbase-hosted demo environment for testing, along with reference implementations and documentation3

3

. The models can navigate familiar interfaces but still struggle with unexpected pop-ups, CAPTCHAs, dynamically loaded content, and unfamiliar layouts2

2

.