Designing Accountable Human–AI Teaming in Industrial Production: A Governance Framework for Agentic, Tool-Mediated Workflows

In industrial production, agentic artificial intelligence (AI) systems combine goal-directed planning with tool-mediated execution and iterative monitoring across the manufacturing execution system, enterprise resource planning, and operational technology layers. Although this autonomy promises responsiveness under volatility, it redistributes decision rights across human supervisors, orchestrator agents, specialist agents, and product owners, turning accountability from a user-interface-level checkpoint into a workflow-level governance problem. Existing responsible-AI and governance research provides valuable principles; however, the following remain underspecified: operationalization of accountability, escalation and override rights, and minimally sufficient traceability in multi-agent production workflows. This study develops a conceptual accountability framework for human–agent–agent cooperation in industrial production. Drawing on a targeted synthesis of information systems and operations management literature, the framework specifies (i) core accountability dimensions for agentic workflows, (ii) a decision-class-anchored responsibility allocation inspired by the RACI framework, and (iii) a minimal evidence bundle for traceability and incident reconstruction, complemented by lifecycle assurance requirements. The framework is applied to a representative high-mix production scenario and refined through semi-structured expert interviews with senior practitioners from DAX-40 automotive manufacturers, a US-based software corporation, and globally leading consulting firms. The study provides actionable design guidelines for developing accountable agentic production systems. It reconceptualizes accountability as a workflow-level governance property, shifts algorithmic accountability from model-level attribution to decision-class-anchored responsibility allocation, and offers a compact design blueprint — comprising risk-calibrated control points, minimally sufficient traceability, and lifecycle governance — for deploying agentic AI in industrial production.