The convergence of Artificial Intelligence (AI) and Circular Economy (CE) is often suggested as a promising pathway for sustainable transformation in the agri-food sector. Yet, how human involvement is configured across AI systems, and whether it enables or limits circular outcomes, remains poorly understood. This study examines how AI is integrated into CE-related practices in agri-food organisations and how human roles are distributed across the AI lifecycle, with particular attention to organisational size and structure. Based on a systematic review of 88 peer-reviewed articles published between 2015 and 2025, the study offers a cross-cutting analysis of how AI systems are conceptualised, developed, deployed, and monitored, and how CE goals are addressed within these processes. Results show a strong imbalance: the early phases of the AI lifecycle are largely shaped by external technology providers, while internal organisational actors are mostly engaged during implementation and operational tasks. Strategic and governance-related phases remain underexplored, and participatory approaches are rare. CE considerations are often reduced to operational metrics, such as waste reduction or resource efficiency, rather than driving systemic redesign or regenerative business models. These findings highlight the need to reposition human involvement as a central lever in aligning AI with circular outcomes. Without stronger organisational capacity, inclusive governance, and CE-aware design logics, AI risks reinforcing existing inefficiencies instead of enabling ecological transition. The study offers a timely contribution to debates on digital sustainability and agri-food innovation, proposing future research directions that foreground agency, adaptability, and systemic value creation.