AI-assisted software development has made implementation nearly costless, exposingspecification and value-verification as the binding constraints on organizational throughput.This paper identifies a further structural limit: the novelty wall. Syntheticusers—AI-generated models of user behavior trained on historical interaction data—caninterpolate within existing value spaces but cannot extrapolate to genuinely novel values.This limitation is not a current capability gap but a structural consequence of statisticallearning from past distributions. We formalize this as the interpolation boundary: anyvalue-verification method that relies on models trained on historical data is structurally blindto values that have no precedent in the training distribution. We then observe that even amonghumans, the capacity to recognize novel value is rare—explaining the persistent socialfunction of value pioneers (influencers, fashion leaders, taste-makers) whose role is notauthority-based but value-delivery-based: they experience, recognize, and translate emergentvalue into forms others can perceive. This analysis extends the Behavior Space Model's Axis2 (specification-against-value verification) by demonstrating that the value-verification looprequires not merely human experience but a specific, rare human capacity for novel valuerecognition—a capacity that is structurally irreducible to computation. Three implicationsfollow: (1) the value-verification loop cannot be closed by synthetic users; (2) Bainbridge'sirony of automation applies recursively to the specification domain itself; (3) the socialinfrastructure for novel value recognition is a non-automatable organizational asset. A fourth,broader implication: wholesale displacement of human employment is not merely sociallycostly but economically self-defeating—an economy restricted to interpolation withinexisting value spaces has foreclosed its own capacity for innovation.
Franny Philos Sophia (Sat,) studied this question.