Blab Tech
Google Adds Computer-Use Capability to Gemini 3.5 Flash, Enabling AI Agents to Interact with Software Interfaces
On June 24 2026, Google announced that its Gemini 3.5 Flash model now includes a Computer‑Use tool, a first for the company’s flagship generative‑AI platform. The new feature lets Gemini‑powered agents read screenshots of graphical user interfaces and perform actions—clicking buttons, filling out forms, navigating web pages—just as a human user would.
The addition moves Gemini 3.5 Flash beyond text‑only conversation. Developers can now build agents that interact directly with existing applications—web browsers, desktop programs, mobile apps—without writing custom APIs or integration layers. The model interprets visual content, plans a sequence of actions, executes them, and evaluates the outcome before proceeding, giving agents a full loop of perception, planning, and execution.
Computer‑Use is aimed at long‑horizon tasks that require planning, decision‑making, and multiple steps. Google says the feature is optimized for the “agentic era,” where AI systems autonomously complete workflows that span several applications. The capability is available in the public preview of Gemini 3.5 Flash and can be accessed through Google’s AI tools, including Vertex AI and the open‑source Agent Development Kit.
Enterprises stand to benefit from several concrete use cases. Customer‑support agents could navigate internal dashboards, retrieve records, and update tickets without manual intervention. Back‑office staff could automate repetitive data‑entry tasks across legacy systems. Software developers could employ AI agents to test applications by interacting with them as a real user would. Researchers could let agents browse websites, collect data, and generate summaries with minimal supervision.
Gemini 3.5 Flash itself is positioned as a high‑performance model that balances speed and intelligence. According to Google, it delivers frontier‑level reasoning, coding ability, and tool usage while maintaining the low latency and cost efficiency of the Flash series. The model supports large context windows, enabling it to process extensive codebases or long documents in a single prompt.
The announcement comes amid a broader push toward autonomous digital agents. Other AI companies are also developing systems that can perform meaningful work on behalf of users, moving beyond conversational chatbots. By enabling direct interaction with software that was not designed for machine automation, Google’s Computer‑Use feature could reduce the need for costly custom integrations.
Google’s release notes for June 24 2026 state that the Computer‑Use tool is available in public preview and includes simplified actions with intents, as well as built‑in support for browser, mobile, and desktop environments. The feature is intended to help developers create reliable agents that can see, reason, and act across multiple platforms.
At present, the capability is still in preview. Developers can experiment with the tool through the Gemini API and the Agent Development Kit, but production use will likely require further testing and validation. Google has not announced a commercial rollout date.
In summary, Google’s addition of Computer‑Use to Gemini 3.5 Flash marks a significant step toward fully autonomous AI agents that can operate within existing software ecosystems. The feature promises to streamline enterprise workflows, reduce manual labor, and accelerate the adoption of generative AI in practical applications. The industry will watch how quickly developers adopt the tool and how it competes with similar offerings from other AI vendors.
The addition moves Gemini 3.5 Flash beyond text‑only conversation. Developers can now build agents that interact directly with existing applications—web browsers, desktop programs, mobile apps—without writing custom APIs or integration layers. The model interprets visual content, plans a sequence of actions, executes them, and evaluates the outcome before proceeding, giving agents a full loop of perception, planning, and execution.
Computer‑Use is aimed at long‑horizon tasks that require planning, decision‑making, and multiple steps. Google says the feature is optimized for the “agentic era,” where AI systems autonomously complete workflows that span several applications. The capability is available in the public preview of Gemini 3.5 Flash and can be accessed through Google’s AI tools, including Vertex AI and the open‑source Agent Development Kit.
Enterprises stand to benefit from several concrete use cases. Customer‑support agents could navigate internal dashboards, retrieve records, and update tickets without manual intervention. Back‑office staff could automate repetitive data‑entry tasks across legacy systems. Software developers could employ AI agents to test applications by interacting with them as a real user would. Researchers could let agents browse websites, collect data, and generate summaries with minimal supervision.
Gemini 3.5 Flash itself is positioned as a high‑performance model that balances speed and intelligence. According to Google, it delivers frontier‑level reasoning, coding ability, and tool usage while maintaining the low latency and cost efficiency of the Flash series. The model supports large context windows, enabling it to process extensive codebases or long documents in a single prompt.
The announcement comes amid a broader push toward autonomous digital agents. Other AI companies are also developing systems that can perform meaningful work on behalf of users, moving beyond conversational chatbots. By enabling direct interaction with software that was not designed for machine automation, Google’s Computer‑Use feature could reduce the need for costly custom integrations.
Google’s release notes for June 24 2026 state that the Computer‑Use tool is available in public preview and includes simplified actions with intents, as well as built‑in support for browser, mobile, and desktop environments. The feature is intended to help developers create reliable agents that can see, reason, and act across multiple platforms.
At present, the capability is still in preview. Developers can experiment with the tool through the Gemini API and the Agent Development Kit, but production use will likely require further testing and validation. Google has not announced a commercial rollout date.
In summary, Google’s addition of Computer‑Use to Gemini 3.5 Flash marks a significant step toward fully autonomous AI agents that can operate within existing software ecosystems. The feature promises to streamline enterprise workflows, reduce manual labor, and accelerate the adoption of generative AI in practical applications. The industry will watch how quickly developers adopt the tool and how it competes with similar offerings from other AI vendors.
