Recent research has shown that currently available AI shopping agents still face significant challenges in making complex purchasing decisions on behalf of users and securely completing transactions. While the vision of fully autonomous AI shopping is compelling, substantial technical constraints remain.
This study evaluated the performance of multiple AI shopping agents across actual purchasing workflows. The evaluation revealed that while AI demonstrates a certain level of competence in simple information gathering and product comparison, numerous errors and uncertainties arise during practical stages such as payment processing, individualized condition negotiation, and ensuring transaction reliability.
Although current AI models exhibit high performance in logical reasoning and natural language processing, the study concludes that they are still underdevelopment when it comes to meeting the rigorous form-input requirements of real-world e-commerce sites, adapting to dynamic price fluctuations, and exercising flexible judgment in unforeseen circumstances.
The research team points out that for agents to engage in fully autonomous shopping, more robust error-handling capabilities and improved control technologies to accurately reflect user intent are indispensable. Moving forward, overcoming these technical barriers will be a critical challenge for the entire industry.