AI coding creates two kinds of debt. You’re only measuring one

A new concept called 'cognitive debt' describes the accumulated loss of understanding of what a system does and why — distinct from technical debt. Coined by Dr. Margaret-Anne Storey, it emerges when AI generates code faster than engineers can build mental models of it. Infobip's longitudinal study of 225 interns found that AI-era cohorts showed significantly lower technical skill growth (+1.56 vs +2.50) compared to pre-AI cohorts, even as satisfaction scores and NPS rose. The key differentiator wasn't AI usage frequency but cognitive engagement: interns who understood and questioned AI output grew, while those who passively accepted it stagnated. Practical countermeasures include 'try first, AI second' habits, AI-free checkpoints, measuring skill trajectory over satisfaction, and team retrospectives that explicitly surface shared understanding gaps.