OpenAI has introduced a new product called ChatGPT Agent, a tool designed to complete complex, multi-step tasks on behalf of users by controlling an entire computer environment. The feature goes beyond traditional chatbots by allowing AI to interact with calendars, browsers, and other software to perform various real-world tasks.
The new tool was announced on Thursday during a demonstration to The Verge by OpenAI’s product lead Yash Kumar and research lead Isa Fulford. They explained that ChatGPT Agent can handle tasks like scheduling meetings, planning meals and shopping for ingredients, preparing presentations, and even generating in-depth research reports.
OpenAI said the agent operates using a specially trained model that combines technologies from two of its previous tools, Operator and Deep Research. The model was developed with reinforcement learning and can use multiple tools, including a text browser, visual browser, and a terminal that lets users import their own data.
In practice, the agent can connect to services like Google Calendar to plan events and cross-check OpenTable for restaurant reservations. Users can adjust the task midway through the process, such as adding a new restaurant category, and the AI will adapt.
OpenAI employees have already started using the tool for everyday tasks. For example, Kumar said he uses it to automate weekly parking requests at the company’s San Francisco office.
The tool is currently slower than real-time interaction. However, Fulford explained that the goal is to tackle difficult tasks, even if they take 15 to 30 minutes. She noted that users can start a task, leave it running in the background, and return later to find it completed. The AI pauses to ask for user approval before taking actions like sending emails or making bookings.
OpenAI also stated that it has implemented strict safety measures, especially concerning potential misuse in biology or chemistry, although the company said there is no evidence the tool could cause direct harm in these areas.
Financial transactions are currently restricted, and ChatGPT Agent includes a Watch Mode for extra security. For example, if a user tries to navigate away from the tab while visiting financial websites, the agent will stop working.
Starting today, ChatGPT Agent will be available to Pro, Plus, and Team users. Access is provided through an “agent mode” option in the tools menu or by typing “/agent.” OpenAI plans to roll it out to Enterprise and Education users later this summer. There is no set date yet for availability in the European Economic Area and Switzerland.
The launch of ChatGPT Agent reflects a growing trend in the tech industry toward developing AI agents—tools that can go beyond conversation and actually complete tasks. Companies like Amazon, Google, Meta, and Klarna have all invested heavily in this technology over the past two years. For example, Klarna revealed in early 2024 that its AI agent handled two-thirds of its customer service chats in just one month.
OpenAI previously introduced Operator in January 2025, which allowed AI to interact with web pages. Other companies, such as Anthropic and Perplexity, have also released similar tools.
With ChatGPT Agent, OpenAI is moving closer to the long-term vision of a digital assistant similar to Iron Man’s J.A.R.V.I.S., capable of managing a range of personal and professional tasks in the background.
