OpenAI has recently unveiled a new demo showcasing its next-generation ChatGPT application's capability to interact directly with desktop applications and utilize an integrated browser to automate research tasks. This feature marks a significant shift from a purely text-based chatbot to an action-oriented AI agent.
Key Highlights
In the latest demonstration video released in July 2026, OpenAI developer Dominik Kundel performed real-time control operations. According to the official ChatGPT X account, the tool can now not only answer questions but also accompany users in navigating the operating system. The AI can proactively open installed software on personal computers, search for information, and execute complex tasks in real time without requiring continuous manual human intervention.
Technical & Technological Analysis
The new feature operates on a 'computer use' mechanism combined with a browser deeply integrated into the core of the ChatGPT application. Unlike standard web extensions, this built-in browser allows the AI model to decode a webpage's DOM structure, automatically click, fill out forms, and scroll like a real user. The system also continuously performs screen parsing to understand the visual context of third-party applications running on the operating system, thereby determining the next course of action.
Expert Opinions & Insights
Tech observers note that OpenAI's integration of direct computer control into ChatGPT is a strong response to competitors like Anthropic's Claude (which previously launched its 'Computer Use' feature). This integration simplifies workflows as users no longer need to copy-paste data back and forth between the chatbot and other productivity tools. However, security experts also warn of potential information security risks when granting an AI the authority to control the mouse and keyboard on personal devices.
Impact & Future Outlook
The arrival of ChatGPT with computer control capabilities promises to transform office workflows for users globally, including in Vietnam. Instead of manually performing repetitive tasks like data entry, report reconciliation, or web scraping, users can fully delegate these to the AI as a digital assistant. This is a crucial stepping stone toward the era of Agentic AI—artificial intelligence entities capable of operating autonomously and solving complex end-to-end problems.