When Agentic AI Participates in Organisational Change: Rethinking Agency, Leadership, and Accountability

Leadership and organisational change have long been theorized as configurations of human actors, each of whom can in principle be held responsible for their part. Agentic AI unsettles this assumption. We argue that such systems increasingly shape what organisational actors see, compare, and decide within change processes, exercising real influence while remaining incapable of bearing responsibility for it. We reframe agentic AI as a consequential but non-responsible participant in organisational change, and we introduce human-AI constellations to describe how practical influence is configured, and continually reconfigured, across AI-dominant, human-dominant, and interdependent arrangements. We treat agentic AI as the sharpest case within a broader family of AI modes that also includes predictive and generative systems, and we distinguish our account from shared leadership approaches that presuppose mutual recognition. We identify three conceptual openings for change scholarship, concerning power without intent, trust without reciprocity, and accountability and ethics without a responsible subject. These shifts reconceive leadership itself, from coordinating human actors towards setting direction, sustaining trust, and legitimating change when one participant is consequential yet stands outside the moral order the others inhabit.MAD statementThis leading article aims to Make a Difference (MAD) by challenging the human-centred assumptions that continue to inform much of change and leadership scholarship. We conceptualize agentic AI as a consequential but non-responsible participant, a system capable of structuring decision spaces and shaping organisational action while remaining categorically incapable of bearing responsibility for what it influences. We introduce human-AI constellations to distinguish how practical influence is configured, and continually reconfigured, across AI-dominant, human-dominant, and interdependent arrangements, and we show how these configurations can align or fail to align actual control with formal accountability. In doing so, we reconceive leadership in change as the work of setting direction, sustaining trust, and legitimating decisions when one participant is consequential yet cannot be held to account, and we open a research agenda on power, trust, accountability and ethics in AI-mediated organisational change.