Multi-agent AI

Abstract Multi-agent artificial intelligence (MAAI) represents a foundational shift in the automation of knowledge work, moving beyond static workflows toward adaptive systems of interacting AI-based agents. These agents perceive, reason, and coordinate in real time to address complex, context-rich tasks that traditionally require human expertise. Drawing on the conceptual roots of process automation, agentic information systems, and AI, this paper introduces a structured, five-component framework that conceptualizes MAAI as a layered architecture composed of foundation model, data-centric perception and action, dynamic orchestration, agent-integrated workflow, and interaction interface. This framework disentangles the technical, organizational, and human-facing dimensions of MAAI, offering researchers and practitioners a systematic lens to analyze and design agent-based AI automation. The framework further structures three research pathways focused on advancing technical capabilities, enabling organizational integration, and addressing socio-technical implications such as fairness, accountability, and labor transformation. Together, these contributions establish a foundation for interdisciplinary inquiry into how MAAI reshapes work, coordination, and digital value creation.