IAGD-PUBL - AI Governance and Cognitive Debt in Software Systems: Empirical Analysis of Production–Comprehension Imbalance

This study investigates the systemic impact of large-scale integration of artificial intelligence in software development processes. It introduces the concept of cognitive debt, defined as the cumulative imbalance between AI-assisted production capacity and human comprehension capacity. A formal model is proposed to describe the dynamic interaction between production, understanding, auditability, and resulting software quality. Numerical simulations and empirical validation on real-world GitHub repositories (Django, FastAPI, Hugging Face Transformers, Kubernetes) demonstrate the existence of stable, critical, and divergent regimes. The results show that excessive reliance on AI-driven production leads to a measurable degradation of auditability and system quality. The study highlights the necessity of maintaining a balance between production and comprehension, summarized by the principle: "Create less, but create better." This work contributes to the fields of software engineering, AI governance, and systemic risk analysis. Scientific Contact:Frédéric CordierEmail: cordierfr@wanadoo.fr or frederic.cordier@alcahest-quantique.frORCID: 0009-0001-8661-3275Project page: https://www.alcahest-quantique.fr