In a significant leap toward realizing artificial general intelligence (AGI), OpenAI has unveiled ChatGPT Agent, a groundbreaking tool that enables its AI to perform complex tasks autonomously using a virtual computer. OpenAI CEO Sam Altman marked the occasion with a bold statement on X: “You can feel the AGI.”

The ChatGPT Agent seamlessly merges conversational intelligence with task automation, web browsing, coding, and app integration, allowing users to offload real-world jobs—like scanning calendars, summarizing meetings, researching competitors, shopping online, or even submitting expenses—with minimal guidance. It works through a self-contained virtual machine, preserving context while switching fluidly between browsing and execution.

This new feature is currently available for ChatGPT Pro, Plus, and Team users in the United States, with Enterprise and Education versions expected soon. Pro subscribers will receive 400 agent-mode messages per month, while others will have access to 40. For those needing more, a credit-based add-on system is available. Deep Research, one of the tools bundled into the agent’s suite, will remain separately accessible for users preferring more detailed outputs.

Unlike its predecessor, Operator—limited to website interaction—or the slower, analytical Deep Research mode, the ChatGPT Agent brings both capabilities together in one cohesive experience. It also integrates with apps like Gmail and GitHub, expanding the reach of what it can do. Whether it’s planning a weekly report or crafting a project presentation, the agent uses a mix of tools—visual browser, terminal, text input, and more—to complete multi-step workflows.

However, OpenAI emphasizes user control and safety. The AI won’t take critical actions—like logging into accounts, sending emails, or making online purchases—without first seeking explicit approval. Users can pause, cancel, or step in during any task. There’s also an option to schedule tasks to run regularly, such as generating weekly performance summaries.

With this increased autonomy comes greater responsibility and risk. One major concern is prompt injection—where malicious instructions on a webpage could trick the AI into taking unintended actions. OpenAI says its system is trained to detect and avoid such threats, and actively monitors for anomalies during operation.

Privacy is also a priority. According to OpenAI, ChatGPT Agent doesn’t save login credentials. Users can delete browsing data or log out from all connected sites in a single step. Session data remains confidential, and cookie usage complies with individual site policies.

While the feature is still maturing—especially in areas like slide deck creation, which remains in beta—OpenAI promises ongoing improvements to layout consistency and export quality. The company has yet to confirm when the feature might become available in India or other regions outside the U.S., although deployment in the European Economic Area and Switzerland is reportedly underway.

OpenAI also announced that its earlier Operator preview will be phased out in the coming weeks, giving way to this more advanced and versatile agentic system.