The creation-to-verification shift is the term Annie Vella and Kelly Blincoe use, in The Impact of AI Coding Assistants on Software Engineering: A Longitudinal Study, for a change in where software engineers using AI coding assistants perceive their effort going: the relative balance moves away from creation tasks and toward verification tasks. Their conclusion is that development work is being reorganised rather than simply accelerated.
Usage
The study groups six core development tasks into two theoretically motivated sets based on the Vee model: creation tasks (designing, writing and refactoring code) and verification tasks (reviewing, testing and debugging). Across two questionnaires six months apart, most engineers perceived spending less time on almost all of these tasks, with writing code falling furthest — 82% reported spending less time on it. Among the 88 matched participants, neither group changed significantly in isolation, but the balance between them (verification minus creation) shifted significantly toward verification, with a moderate effect size; decreased creation combined with increased verification was the most common individual pattern.
The authors note that the shift is modest in absolute terms: reviewing code was the only task whose mean sat above neutral, and testing remained below neutral throughout, though it trended upward. Since the increase in verification did not account for the reduction in creation, they propose that the remaining effort flows into Supervisory Engineering Work — directing, evaluating and correcting AI output — which engineers may not recognise as traditional testing or code review.
As implications, the authors suggest that engineers prepare for less time writing code and more time directing and evaluating AI output, and that educators teach judgement, trust calibration and the components of supervisory work alongside traditional technical skills.