OpenAI has launched a major update to ChatGPT with the introduction of the ChatGPT Agent—a feature designed to carry out complex, multi-step tasks online with minimal user input. This agent combines ChatGPT’s advanced conversational capabilities with code execution and web interaction, enabling it to perform tasks end-to-end. The rollout begins with Pro, Plus, and Team users in the US, while Enterprise and Education access is expected soon. Pro users will receive 400 messages monthly, while other paid tiers get 40 messages, with extra usage available via a credit system.
“You can feel the AGI,” OpenAI CEO Sam Altman posted on X, hinting at the broader ambition behind the feature—moving closer to artificial general intelligence, where AI can think and learn like humans.
The new agent builds upon previous tools like Operator and Deep Research but offers a more integrated, autonomous experience. It can do things like scan calendars, summarize meetings, research competitors, create slide decks, and even shop for groceries or file expense reports. It accomplishes this through tools such as a visual browser, terminal access, and integrations with apps like Gmail and GitHub. A key advantage is its use of a virtual machine, which allows it to maintain context across multiple tools and tasks seamlessly.
Importantly, users retain full control. ChatGPT will always ask for permission before performing critical actions like logging in to websites, sending emails, or making purchases. Tasks can be paused, canceled, or scheduled to repeat, such as generating weekly performance reports.
OpenAI acknowledges that greater access brings new risks. A major concern is prompt injection—where hidden prompts on websites could manipulate the agent’s behavior. To counter this, OpenAI has trained the system to detect and resist such attacks and monitors it during sensitive tasks.
Privacy remains a priority: login credentials are not stored, browsing data can be deleted with one click, and sessions follow site-specific cookie policies.
Although powerful, the agent is still evolving. Features like slide deck generation are in beta and may yield inconsistent results, especially from scratch. OpenAI is working to improve formatting and export quality in future updates.
Access in the European Economic Area, Switzerland, and other regions like India remains pending. Meanwhile, the older Operator preview will be phased out in the coming weeks.
