OpenAI introduced a new tool, ChatGPT Agent, which allows users to delegate complex tasks to AI within their own virtual computer. Users can schedule meetings, book restaurant tables, create presentations based on competitor analysis, make online purchases, or automate routine tasks — all through natural language queries without special commands. ChatGPT Agent combines the capabilities of previous OpenAI tools — «Operator» for website interaction and «Deep Research» for in-depth information analysis.
Users can select «agent mode» in the ChatGPT menu or enter «/agent» to access the new functionality. The service works with various interfaces — images, text, terminal — and can connect to Gmail, GitHub, calendars, and other services. When performing tasks, the agent independently selects the necessary tools and allows interruptions or changes in real-time. Before any irreversible actions, such as sending an email or booking, ChatGPT Agent always requests user confirmation.
OpenAI has implemented additional security systems, as the new model has extended capabilities. Access to financial transactions is limited, and for sensitive actions, a «Watch Mode» is provided, where the user must remain in the corresponding tab. All agent actions are subject to monitoring, and a separate switch can delete browsing history.
ChatGPT Agent is already available for Pro, Plus, and Team users, who can try the new features immediately after the update, and soon the tool will be opened for corporate and educational clients. The launch for the European Economic Area and Switzerland is still being prepared. According to initial users, the agent helps save time on creating presentations, preparing reports, and processing mail, although complex multi-step processes can take up to half an hour.
