Is Being Smart Not Enough? MEOW as a Simonian Decision Architecture for Generative AI in Organizations

Abstract

Generative AI lowers the cost of generating decision-relevant summaries, alternatives, and arguments, but expanded model capability does not by itself yield more rational organizational decisions. This conceptual position paper argues that generative AI can redistribute bounded rationality: when generated outputs expand faster than the capacity to assess and use them, binding constraints shift toward verification, attention, decision-premise governance, and learning. Building on Herbert A. Simon and research on human–AI collaboration, AI governance, and organizational routines, the paper proposes MEOW: Model Intelligence, Epistemic Verification, Organizational Decision Premises, and Work Routines. These dimensions describe complementary capabilities enacted within an AI-mediated decision episode, rather than stages in a linear pipeline. Organizational rationality is defined here procedurally as evidence-grounded, premise-explicit, accountable, and revisable decision-making; it does not imply optimal outcomes or ethical legitimacy by managerial authorization alone. Five propositions address generation–verification imbalance, premature satisficing, capability complementarity, risk-calibrated verification, and trace-enabled learning. The framework translates these relationships into roles, evidence thresholds, decision gates, records, and feedback loops. Its proposed contribution is a Simonian integration centered on how AI outputs become accepted decision premises, with observable conditions under which that passage may succeed or fail. It does not claim priority for verification, oversight, or governance gates. The propositions and accompanying decision canvas, checklist, and log remain unvalidated and are offered for empirical testing and context-sensitive organizational design.

Keywords