OpenAI has released a new tool named ChatGPT Agent, designed to autonomously perform complex, multi-step tasks for users by operating a virtual computer. According to reports by Hayden Field for The Verge and Reece Rogers for Wired, the agent can handle a wide range of activities, from personal planning to professional work.
The company says the new tool can perform tasks like analyzing a user’s calendar to plan a date night, which includes finding restaurant availability on third-party websites. It can also generate research reports, create PowerPoint presentations based on its analysis, and fill out forms online. In a demonstration, OpenAI product lead Yash Kumar and research lead Isa Fulford showed how the agent could be interrupted and redirected by the user.
ChatGPT Agent is powered by a new, unnamed AI model that combines the capabilities of two previous OpenAI tools: Operator, which visually navigates websites, and Deep Research, which processes large amounts of text for analysis. Isa Fulford told The Verge that the teams behind both tools were merged to develop the new product. The model was trained using reinforcement learning to use a set of tools that includes a visual browser, a text browser, and a terminal for data.
The company states that the agent is not designed for speed but for completing difficult tasks. A simple request might take five minutes, while generating a research-based slide deck could take around 25 minutes. “Even if it takes 15 minutes, half an hour, it’s quite a big speed-up compared to how long it would take you to do it,” Fulford explained to The Verge. The intention is for users to assign a task and let the agent work in the background.
To ensure user control, the agent must ask for permission before performing irreversible actions like sending an email or making a booking. A feature called “Watch Mode” requires the user to remain on the browser tab when the agent accesses sensitive websites, such as financial portals. Financial transactions are restricted for now. The company has also activated advanced safeguards for potentially harmful capabilities, although it stated there is no evidence the model could help create biological or chemical weapons.
The ChatGPT Agent is being rolled out to paying subscribers of ChatGPT Plus, Pro, and Team, with Enterprise and Education users expected to get access later in the summer. The launch positions OpenAI in the growing market for AI agents, a major trend in the technology industry where companies aim to create assistants that can proactively complete tasks for users.
