The invisible hand of generative AI: quasi-agency and knowledge transformation in data-driven decision-making

Abstract

This study aims to examine how generative artificial intelligence (GenAI) reshapes knowledge management (KM) in data-driven decision-making (DDDM), reconfigures principal–agent relations and creates new challenges for epistemic governance. It focuses on how GenAI influences the construction, synthesis, justification and retention of decision-relevant knowledge across the decision cycle. This study adopts an abductive qualitative design, drawing on data from 120 managers at Italian small and medium-sized enterprises actively using GenAI in recurring DDDM activities. Drawing on ten open-ended questions aligned with the decision cycle, the analysis applies reflexive thematic analysis, supported by Gioia-informed coding procedures, to identify recurring phase-level governance patterns and to develop a phase-sensitive framework. This study develops a phase-sensitive model of GenAI quasi-agency showing how generative artificial intelligence reshapes organizational knowledge throughout the decision cycle. The model explains that GenAI transforms knowledge through three recurring mechanisms – knowledge coupling, knowledge decoupling and knowledge hiding – which operate with different intensity across decision phases and redefine the governance requirements for traceability, reconstructability, contestability and human justificatory ownership. Consequently, agency risks increasingly originate from the transformation of hidden knowledge rather than from hidden human action. This study introduces the concept of GenAI quasi-agency, defined as the capacity of GenAI systems to shape framing, evidence construction, prioritization, justification and organizational memory without possessing formal authority, intentionality or accountability. It contributes to KM by conceptualizing GenAI-supported decision-making as a problem of epistemic governance centered on preserving traceability, reconstructability, contestability and human justificatory ownership. The study also extends principal–agent theory by showing how agency risks increasingly arise from the transformation of hidden knowledge rather than from hidden action alone and develops a phase-sensitive framework that explains how governance requirements vary across the DDDM cycle.