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

arXiv·medium signal

A 13-month causal study of 151 Java repos (74 adopting agentic AI, 77 controls) over 1,811 monthly snapshots finds architectural smell counts essentially flat (+1.1%) while lines of code jumped +12.8%, producing a misleading 6.7% drop in smell density — a denominator effect, not real architectural improvement. The takeaway: density-normalized quality metrics can deceive when AI inflates system size.