OpenAI has unveiled “ChatGPT agent,” an upgrade that enables the AI assistant to autonomously complete real-world tasks using virtual tools, including a browser, terminal, and API access. The new system can independently manage complex workflows like analyzing inboxes, preparing meeting materials, creating presentations from competitor research, and planning a multi-course dinner shopping.
The agent navigates websites both visually and textually, completes forms, accesses authorized accounts, executes code, handles files, and produces editable documents. The upgrade combines features from previous tools, merging Operator’s web navigation capabilities with deep research functions into a unified experience. The agent can alternate between reasoning and action while selecting optimal tools for each task.
ChatGPT agent always requests permission before performing any action with real-world consequences, OpenAI stated. Users maintain control with the ability to pause, intervene, or take over tasks as needed. When account access is required, users retain full session control with private inputs. The feature is initially available to ChatGPT Pro, Plus, and Team subscribers through the tools dropdown menu in conversations, with Enterprise and Education customers gaining access in the upcoming weeks.
