The paradox of efficiency: institutional interfaces, residuals, and the erosion of innovation drivers in AI-augmented organizations.

As artificial intelligence (AI) becomes deeply embedded in organizational cognition and decision-making, a profound paradox emerges: AI significantly enhances operational efficiency while systematically eroding the generation and retention of innovation drivers. This study introduces the institutional interface as a meso-level analytical lens for examining how specific configurations of AI design choices, human-AI interaction protocols, and organizational norms produce systematic biases. Under the design conditions theorized in this article, these configurations tend to erode innovation drivers through unconscious automatic execution and residual production. This study illuminates a key feature of AI-driven institutional execution-unconscious automatic execution-and deconstructs it into three dimensions: opacity of execution, loss of agency, and diminished reflexivity. It introduces the concept of residuals to capture the cognitive, behavioral, and value-laden elements that deviate from statistical mainstreams or resist quantification and are systematically deprioritized or excluded from organizational awareness. Under specific design conditions and viewed through the three nested interface lenses, AI systems tend to filter residuals across the cognitive, action, and value interfaces, systematically excluding potentially valuable innovations and impoverishing organizational cognitive diversity. This study articulates the conditions under which a self-reinforcing tendency toward efficiency lock-in emerges and theorizes the institutional features-non-optimization zones, outlier pathways, and paradox-balancing roles-that may counteract this drift. Critically, it further reveals the persistent and irreducible dilemmas facing such features (isolation, authority, and evaluation), arguing that they are tensions to be continually managed rather than problems to be solved. Advancing testable theoretical propositions, this article offers new analytical tools and design directions for innovation governance in AI-augmented organizations. It contributes directly to discussions on AI's role in business understanding and managerial decision support, while laying the conceptual foundation for a broader research program on the learning and governance of AI-augmented organizations.