Mining Architectural Quality Under Agentic AI Adoption: A Causal Study of Java Repositories

The study investigates the impact of agentic AI adoption on software architecture by analyzing 151 open-source Java repositories, distinguishing between those with AI integration and matched controls. Using a staggered difference-in-differences design, it finds that while architectural smell density (ASD) decreases by 6.7% with a significant increase in lines of code (12.8%), the change is attributed to a denominator effect rather than genuine architectural improvement. This research highlights the necessity for careful metrics in assessing the influence of AI coding tools on architectural quality, providing a publicly available dataset for further exploration.