Dancing with the algorithm: a framework to navigate knowledge and autonomy in AI-assisted managerial decisions

While artificial intelligence (AI) systems (assistive, human-in-the-loop decision support systems) increasingly participate in complex organizational judgments, their integration into knowledge processes raises fundamental challenges for autonomy, trust, and epistemic agency. This study aims to develop a dynamic, phase-specific framework that explains how autonomy evolves in relation to the data–information–knowledge–wisdom (DIKW) hierarchy and foundational knowledge management concepts during AI-assisted managerial decision-making. The study draws on 122 in-depth interviews with senior professionals across diverse sectors, complemented by two expert focus groups. Data were analyzed using the Gioia Methodology to support inductive theory development and generate a grounded conceptual framework. The authors develop the Human–AI autonomy loop (HAIL) framework, mapping decision-making to four recursive phases (frame, evaluate, commit, enact) and DIKW layers, each linked to distinct DIKW layers and autonomy configurations. Autonomy is a situated, distributed practice: managers preserve discretion through interpretive buffers, overrides and moral authorship. Trust in AI is recalibrated by phase, especially as decisions move from information to judgment. HAIL shows that autonomy is sustained through reflexive knowledge practices. This study advances the knowledge management literature by integrating autonomy, trust and epistemic agency into a unified framework of AI-assisted decision-making. It reinterprets the DIKW model not as a linear information hierarchy, but as a socio-technical terrain where knowledge becomes actionable only when embedded in situated judgment and ethical authorship. The HAIL framework offers both theoretical insights and practical guidance for preserving human discretion and organizational wisdom in AI-assisted environments.