OpenAI officially launched the artificial intelligence (AI) feature 'ChatGPT Agent' on the 17th (local time), which handles tasks online on behalf of users and automates complex operations. This feature is designed to allow a single agent to manage a variety of everyday tasks, including document creation, presentation preparation, and restaurant reservations, going beyond simple searches.
The 'ChatGPT Agent' unveiled by OpenAI this time is a product that integrates the previously introduced 'Operator' and 'Deep Research'. The 'Operator' performs execution-based tasks such as booking travel and ordering food on behalf of the user, while 'Deep Research' has the functionality to gather information from various websites and create professional analytical reports.
The newly revealed agent combines the advantages of both functions, enabling visual interface operations such as web page exploration, clicking, scrolling, and text input, as well as executing code, generating files, analysis, API calls, and calendar integration.
OpenAI stated, 'The ChatGPT Agent can quickly read and research multiple pages on the web and directly manipulate the actual website interface to perform tasks.' They added, 'It can also handle restaurant reservations according to the user's schedule.'
In a live demonstration, in response to a request to 'create presentation materials on NVIDIA's quarterly performance,' it automatically generated a presentation file by collecting relevant documents and images through web search and utilizing an image generation API. Additionally, it demonstrated automatically making a reservation by linking to Google Calendar in response to the request, 'Please check the schedule between 6 p.m. and 9:30 p.m. from Tuesday to Thursday and make a reservation for either Italian or Korean restaurant on OpenTable.'
The 'ChatGPT Agent' can be used by selecting 'Agent mode' on the ChatGPT website or by typing 'agent.' It is currently being rolled out sequentially to ChatGPT Pro, Plus, and Team subscription users and is available in most countries, excluding the European Economic Area (EEA) and Switzerland.
OpenAI noted that this agent has strong safety measures, restricting high-risk tasks such as financial transactions and legal advice, and ensuring that user consent is requested before handling sensitive tasks.