OpenAI Launches ChatGPT Agent: AI Now Automates Multi-Step Tasks with Its Own Browser
OpenAI has introduced ChatGPT Agent, a newly launched feature designed to enable its AI assistant to independently handle multi-step tasks by operating its own web browser. This upgrade blends functionalities from OpenAI’s previous Operator tool and its Deep Research feature, creating a system that can browse websites, execute code, and generate documents, all while allowing users to remain in control throughout the process.
This development represents OpenAI’s latest step into the realm of agentic AI—technology designed to perform complex, autonomous actions on a user’s behalf. According to OpenAI, the ChatGPT Agent can manage tasks such as researching and purchasing clothing for a specific event, preparing presentation slide decks, organizing meal plans, or updating financial spreadsheets with new information. To complete these tasks, it leverages web browsing, terminal commands, API access, and integration with popular platforms like Gmail and GitHub through what OpenAI calls “ChatGPT Connectors.”
What sets the ChatGPT Agent apart is its transparent operational process. Within the ChatGPT interface, users can view a dedicated window that displays every action the AI takes inside its secure sandbox environment. This sandbox functions as a virtual computer complete with its own browser and operating system connected to the actual internet. However, it operates independently of the user’s personal devices. OpenAI describes this as the AI shifting fluidly between reasoning and action, capable of managing intricate workflows entirely through user instructions.
For tasks with real-world implications, such as making online purchases, ChatGPT Agent incorporates safety protocols that require explicit user authorization before proceeding. Additionally, users have full control to pause, intervene, or completely stop any task at any time. For more sensitive activities, like sending emails, OpenAI includes a Watch Mode, where the system awaits direct user approval before executing actions.
Since the new Agent surpasses the capabilities of the earlier Operator tool, OpenAI plans to maintain the Operator preview site temporarily before eventually phasing it out, signaling a broader transition toward this more advanced and autonomous solution.
