The velocity–understanding gap names a specific hazard of AI-assisted development: output can accelerate far faster than comprehension. A team may double its shipping rate while the share of merged code any human genuinely understands quietly falls.
The gap is invisible on a velocity dashboard — story points, PR counts, and deploy frequency all look healthy. It surfaces later, as slow incident response, brittle refactors, security findings in code nobody reviewed deeply, and onboarding that stalls because the "how" lives only in a model's context window, not the team's heads.
Making the gap visible is the first step to managing it: measure comprehension alongside throughput, and invest in closing the distance where the code is most critical. Cartara's Velocity–Understanding Gap calculator gives teams a first rough read.