ChatGPT Unveils Agent Mode, Expanding AI Automation to Complex Tasks

OpenAI has officially launched "ChatGPT agent," a new capability designed to empower the AI to plan and execute complex, multi-step tasks on behalf of users. Released on July 17, 2025, the new agent mode integrates advanced functionalities such as web browsing, code execution, and direct interaction with external applications like Gmail, Google Drive, and Calendar.

According to a tweet from "The Artificially Intelligent Enterprise," > "ChatGPT just leveled up with Agent Mode. It can now plan tasks, browse, run code, and pull from your Gmail, Drive, and Calendar to deliver full outcomes." This marks a significant shift from a conversational AI to an active agent capable of performing work. The feature is currently available to users on ChatGPT Pro, Plus, and Team plans.

ChatGPT agent unifies the strengths of OpenAI's previous agentic tools, Operator and Deep Research. Operator provided the ability to interact with websites through a graphical user interface (GUI), allowing for clicks, typing, and form filling. Deep Research, conversely, excelled at extensive text-based web browsing, synthesizing information from numerous sources into comprehensive reports. The new agent seamlessly combines these, enabling it to efficiently read web pages and interact with visual elements. It also gains access to a terminal for running code and generating or analyzing files like spreadsheets and presentations, along with API access for integrated services.

This development positions ChatGPT as a more autonomous tool, moving beyond merely answering questions to actively performing tasks. OpenAI states that the model underlying ChatGPT agent demonstrates state-of-the-art performance on various benchmarks, including a 41.6% score on Humanity’s Last Exam and 27.4% on FrontierMath when utilizing its tools. This indicates a substantial leap in its ability to reason and execute.

OpenAI has emphasized safety in the rollout, classifying ChatGPT agent as "high capability" in biological and chemical domains due to its enhanced abilities. To mitigate potential misuse, the company has implemented real-time monitors that analyze prompts and responses for suspicious activity. Additionally, the memory feature, which allows ChatGPT to recall past conversations, has been temporarily disabled for this agent to prevent prompt injection attacks and safeguard sensitive user data. Users are advised to exercise caution and utilize features like "takeover mode" for sensitive inputs.

While the new agent mode promises to automate time-consuming tasks like planning events, conducting research, and generating detailed reports or financial models, its real-world efficacy will be closely monitored. OpenAI plans iterative improvements, aiming for a more robust and widely accessible agentic AI experience in the future.