Modern LLM agents interact with the web through various architectures - from traditional browser automation to API-based approaches. This project provides implementation and evaluation code to systematically compare their effectiveness across 91 realistic e-commerce scenarios.