OpenAI's ChatGPT Agent: A Promising Yet Flawed AI Shopping Assistant

In a significant advancement for artificial intelligence, OpenAI recently launched ChatGPT Agent, a tool designed to assist users in completing complex, multi-step tasks online. Available exclusively through the $200-per-month Pro subscription, this innovative service aims to streamline online shopping and other consumer activities. However, initial user experiences reveal limitations that highlight both the potential and the challenges of this new technology.

The rollout of ChatGPT Agent follows the integration of features from previous OpenAI products, specifically Operator and Deep Research. According to OpenAI’s product lead, Yash Kumar, the agent is intended to operate in the background, allowing users to engage in other activities while it tackles chores like online shopping or planning activities (Field, 2025). Yet, as evidenced by recent tests, the agent's performance has been mixed at best.

Upon activating ChatGPT Agent in what OpenAI terms "Agent Mode," users are presented with a variety of tasks it can undertake. These include finding specific products, creating lists, and managing calendars. However, as experienced by the author of this article, the agent's execution of these tasks can be sluggish and sometimes inaccurate. For instance, when tasked with finding a vintage lamp on Etsy, the agent took nearly 50 minutes to process the request, only to fail to add the selected items to the user's cart effectively (Field, 2025).

This sluggishness is coupled with a lack of direct control over the user's browser, which significantly limits the agent’s functionality. According to OpenAI's research lead, Isa Fulford, while the agent is designed to optimize for difficult tasks, speed is not its primary focus. This raises questions about its practical application in everyday scenarios where quick results are often essential (Field, 2025).

Another notable limitation of ChatGPT Agent is its inability to handle sensitive financial transactions. When prompted to set up automatic transfers or access bank accounts, the system responded with error messages and ultimately clarified that it could only assist with lower-stakes consumer purchases, such as groceries (Field, 2025). This restriction reflects ongoing concerns about security and user privacy, particularly in financial contexts.

The agent's challenges extend to its handling of flower delivery services, as the author discovered during a test to send flowers to a friend. While the agent provided a list of options, it lacked the capability to complete the ordering process directly, requiring users to take manual steps (Field, 2025). This limitation underscores a crucial distinction between an assistant that can gather information and one that can execute tasks autonomously.

Despite these drawbacks, the potential for ChatGPT Agent to assist users is evident. It can provide detailed analyses and recommendations, making it a valuable tool for decision-making, albeit with significant caveats regarding its execution capabilities. OpenAI's ambitions for ChatGPT Agent highlight the evolving landscape of AI and its implications for everyday tasks.

In conclusion, while OpenAI’s ChatGPT Agent represents a progressive step in AI technology, its current limitations in speed and task execution raise important questions about its overall utility. Users looking for an efficient assistant for online shopping and complex tasks may find themselves navigating a frustrating experience, at least for the time being. As OpenAI continues to refine the agent, future improvements may enhance its functionality and reliability, but for now, it serves as a reminder of the challenges inherent in developing effective AI solutions.