OpenAI launches a new ChatGPT agent tool for autonomous multi-step task completion

OpenAI has introcuced ChatGPT agent, a new tool that can autonomously perform multi step tasks on your behalf, such as browsing the web, running code, analyzing data, retrieving documents, and generating presentations. It acts as a digital worker inside a virtual computer, managing full workflows across different tools while retaining context throughout the session.

The agent brings together existing tools in ChatGPT like Operator for interacting with websites like a user, Deep Research for collecting information and writing detailed reports, and Connectors to access services like Gmail, Google Drive, or Dropbox. These components now work together within a single persistent environment that includes a terminal, web browsers, and a file system. These capabilities allow it to complete full workflows such as downloading files, modifying them via terminal, and reviewing results without losing context. For example, it can retrieve a document from Drive, open and analyze its contents, generate a formatted report or presentation, compare products online, manage your calendar, summarize upcoming meetings, or assist with online purchases. The company says task completion time varies, with some taking 15 to 30 minutes.

The rollout of the new tool will begin with ChatGPT Pro users (400 queries/month), followed by Plus and Team users (40 queries/month), and then Enterprise and Education customers. Paid users can access the agent via the tools dropdown in ChatGPT as “agent mode,” with live on-screen narration and manual override capability.