The Design Function Hidden in Code: A Theory of Constraints Analysis of the AI Coding Productivity Paradox

AI coding assistants accelerate individual developers by 20–55%on isolated tasks, yet organizational delivery shows noimprovement or actively declines. Developers report feeling 20%more productive while controlled measurement shows 19% slowerperformance. This paper provides a theoretical framework thatexplains the paradox.We argue that coding, as practiced by human engineers, is acomposite design act: a single cognitive activity thatsimultaneously performs intent definition, specificationconcretization, consistency verification, testing, andimplementation. These functions are inseparable in humanpractice and invisible as separate activities. AI coding agentsdecompose this composite, accelerating implementation tomachine speed while the design functions — which requiredeliberate human judgment — are silently lost.We analyze the consequences through three theoretical lenses.The Behavior Space model characterizes software behavior alongspecification and verification axes, revealing that AI-generatedcode populates the system with unspecified-but-verified behavior(Ev) that creates false confidence. Goldratt’s Theory ofConstraints explains the degradation mechanism: implementation,no longer the binding constraint, overproduces and floodsspecification and verification with unprocessable work-in-process.Gustafson’s Law explains the compounding: faster implementationenables larger projects, expanding specification burden super-linearly.The framework corrects a widespread misattribution — the AIindustry invokes Amdahl’s Law where the Theory of Constraints isthe appropriate model — and explains why engineering expertswere systematically misled by gradual tool evolution, tacitknowledge, and Goodhart-corrupted metrics. Six empirical studiesconfirm the framework’s predictions. The practical implication isdirect: the resolution is not better AI but better constraintmanagement — subordinating implementation rate tospecification capacity.