Beyond Functional Correctness: Design Issues in AI IDE-Generated Large-Scale Projects

This project is essentially an academic research artifact (0 stars, 7 forks) that evaluates the limitations of current AI coding agents in maintaining software design principles. While the insights are valuable for software engineering researchers, it lacks a technical moat or product-market fit. The study's findings are a snapshot in time; as frontier models like Claude 3.5 Sonnet or GPT-4o evolve, the specific design flaws identified (e.g., modularity issues, god-classes) will likely be addressed via better system prompting or fine-tuning by the frontier labs themselves. Frontier labs are highly likely to internalize these evaluation metrics into their RLHF (Reinforcement Learning from Human Feedback) pipelines to improve the 'architectural thinking' of their models. Projects like SWE-bench already provide more robust, functionally-driven benchmarks, and this project fills a niche for qualitative 'clean code' metrics that is easily absorbed by the platforms being studied (Cursor, GitHub Copilot, etc.).