OpenAI has introduced a new feature, ChatGPT Agent, a virtual assistant that can control your computer and perform multi-step tasks. The Agent is now available to ChatGPT Pro, Plus, and Team users, and will launch for enterprise accounts by the end of the summer.
Unlike a typical chatbot, an Agent can manage a calendar, create presentations, analyze documents, make online purchases, schedule meetings, and perform everyday tasks like restaurant reservations. It has access to a virtual environment with a browser, terminal, file system, and other tools, allowing it to handle requests autonomously.
The feature is based on a new model that can interact with multiple tools at once. The development team combined the efforts of the Deep Research and Operator projects, but unlike the latter, Agent does not simply advise or help in making decisions - it performs actions independently in a virtual environment. Operator, on the other hand, involves human participation in completing the task.
ChatGPT Agent can:
- Analyze the calendar and plan events;
- create documents, tables, presentations;
- work with user files;
- search and book through third-party services;
- generate reports and process web data.
The tool requires confirmation before performing irreversible actions, such as sending an email or placing an order. Some features, including transactions, are temporarily limited, and some sites open in Watch Mode.
The feature is not yet available in the European Economic Area and Switzerland. OpenAI has not announced an exact launch date for these regions.
The new tool comes amid a general competition in the AI agent space. Google, Meta, Amazon, Anthropic, and startups like Klarna are already implementing similar solutions. According to OpenAI, ChatGPT Agent is a step towards an AI model that not only advises, but also performs real work.