Agentic AI promises to revolutionize e-commerce by enabling autonomous shopping agents that independently research, compare, negotiate, and purchase on behalf of users – shifting from reactive chatbots to proactive decision-makers. Yet, real-world deployment faces underexplored challenges: ensuring these agents align with diverse user needs, maintain trust during oversight, and outperform simpler LLM-based or traditional e-commerce interfaces. 

Problem Description

Agentic commerce involves AI systems that can perceive user goals (e.g. 'sustainable shoes under €100'), execute multi-step tasks autonomously across retailers and adapt via feedback loops. However, current systems lack validated user requirements for usability, transparency and autonomy in hybrid human–AI scenarios. While existing LLM assistants can handle queries reactively, they struggle with complex autonomy. Fully manual e-commerce, on the other hand, demands excessive user effort. There have been no studies that derive relevant user requirements for agentic commerce, nor are there any empirical prototypes that compare agentic designs against e-commerce and LLM-only benchmarks. This leaves a research gap in the topic intelligent shopping in the information systems discipline.

Goal of the Thesis

This thesis follows a design science research methodology to (1) analyze agentic commerce via literature review and potential user interviews, deriving a set of user requirements; (2) design and implement a functional prototype agent using established frameworks; and (3) evaluate it empirically against an LLM-only baseline (e.g., GPT-based advisor) and a traditional e-commerce interface (e.g., Amazon-like simulation) in a small-scale user study (12-15 participants). Insights will yield design guidelines for scalable agentic systems in e-commerce.

Requirements

• AI expertise: Practical experience with LLMs (e.g. prompt engineering and OpenAI APIs).
• Programming skills: Basic development skills, preferably in Python and web development.
• Methodology: Familiarity with design science research (Hevner et al.) is preferred but not essential.
• Language skills: Proficient English for literature and documentation.