San Francisco, California — OpenAI’s latest innovation, ChatGPT Agent, is designed to assist users with a variety of complex tasks, functioning much like a day-one intern—albeit a somewhat slow one. Officially launched just a day ago, the tool integrates features from the company’s previous offerings, such as Operator and Deep Research, and aims to streamline multi-step processes on behalf of its users.

To access ChatGPT Agent, users can activate “Agent Mode” by simply typing “/agent.” This prompts the AI to suggest a range of tasks, from finding a budget-friendly coffee grinder to planning a date night. One notable example included locating a vintage lamp on Etsy under $200—a task that highlights both the AI’s capabilities and its limitations.

Upon selecting the lamp option, the AI generated a detailed prompt that aimed to clarify user expectations. However, the actual execution demonstrated significant delays. The ChatGPT Agent took around 50 minutes to complete the task, detailing its every move—from loading the Etsy site to filtering search results. Unfortunately, while it claimed to add values to a virtual shopping cart, the items did not appear in the user’s actual cart, revealing a gap in its interface and functionality.

While each step is meticulously documented by the AI, experts and early users note that the sluggishness can hinder its practicality. OpenAI acknowledges that its primary focus is on optimizing the agent for intricate tasks, not necessarily on speed. In a recent demonstration, product and research leads explained the goal is to allow users to initiate tasks in the background while attending to other matters.

Such limitations become pertinent when discussing financial capabilities. ChatGPT Agent will not be handling monetary transactions due to security protocols that prevent it from accessing sensitive financial information. Users cannot delegate banking tasks or initiate transfers; instead, the AI can assist with everyday purchases, such as groceries, while maintaining strict protocols for high-stakes financial operations.

Despite these restrictions, the agent operates adequately within its designated capabilities. When tasked with ordering flowers for a friend, for instance, it successfully retrieved options while advising on delivery sources. However, even in this situation, it fell short by failing to assist directly with the ordering process, thus underlining its status as a guiding assistant rather than an autonomous executor.

One critical observation made during testing was the agent’s reliance on external resources—highlighting both its strengths in research and its weaknesses in execution. While prompted to assist in a purchase, the AI mentioned it couldn’t access websites without direct user input, leading to confusion about its operational scope.

ChatGPT Agent shines in navigating complex queries and suggesting options but falters in executing those options seamlessly. Users may find its ability to write and provide guidance beneficial, yet the lack of functionality to finalize transactions directly may leave many feeling like they are still doing the heavy lifting.

In summary, while ChatGPT Agent represents a notable step in AI advancement, significant room for improvement remains. As OpenAI continues to refine this tool, it will be essential to address these concerns to enhance user experience and deliver on promises made for automated assistance.