OpenAI officially introduced a new artificial intelligence tool on the ChatGPT platform — a feature called ChatGPT Agent. Thanks to this innovation, users will no longer be limited to simply asking questions and receiving answers; they will now be able to delegate specific tasks directly to the agent.
The new agent is capable of performing various computer-based tasks on behalf of the user. For instance, it can prepare presentations, write code, conduct research online, or plan meetings by checking the user’s calendar. This marks a significant step forward in transforming ChatGPT from a conversational AI into a digital assistant that can actually carry out tasks.
ChatGPT Agent combines the capabilities of several tools previously introduced by OpenAI. For example, the “Operator” tool allowed users to navigate and click through web pages, while “Deep Research” could analyze multiple sources and generate clear, concise reports. These functionalities have now been merged and enhanced within a single, more powerful agent.
Users can interact with the agent in a very natural and intuitive way — simply by typing in everyday language. For example, one might write: “Plan a Japanese breakfast for four people and buy the necessary ingredients.” The agent will then search the web, identify what’s needed, and even show where and how to purchase the items. It can also handle more complex tasks, such as: “Analyze my competitors and prepare a slide deck with the findings.”
These features are currently available only to subscribers of OpenAI’s Pro, Plus, and Team plans. To activate the agent, users need to select the “agent mode” from the tools menu within ChatGPT.
OpenAI states that the model behind this agent is significantly more capable than its predecessors. For example, in the challenging test called Humanity’s Last Exam, the model achieved a score of 41.6%, nearly double that of previous models. In an even more difficult benchmark, FrontierMath, the agent scored 27.4%, whereas the best previous result was only 6.3%.
OpenAI has also emphasized the importance of safety with this release. The company notes that due to the model’s potential for misuse in biological or chemical contexts, additional safeguards have been put in place. Every user prompt is monitored in real time. If a query is related to biology, it undergoes an extra layer of scrutiny. Additionally, the memory function for this agent has been temporarily disabled to prevent misuse and ensure user data remains secure.
The concept of AI agents is currently a hot topic in the tech world, with many companies introducing similar products. However, most of these agents still struggle to handle complex, real-world tasks. OpenAI, on the other hand, claims to have developed a more agile, powerful, and practical solution with ChatGPT Agent.
Time will tell how useful this new agent proves to be in daily workflows. But one thing is certain: AI no longer just talks — it works.
