Explainable Action Chains in Agentic AI: Enhancing Transparency and Trust

This study addresses the growing challenge of explainability in Agentic Artificial Intelligence (Agentic AI), where autonomous systems operate through sequential, goal-driven decision processes rather than isolated predictions. The primary objective of this study is to develop and experimentally evaluate a behavior-centric explainability framework, termed Explainable Action Chains (XAC), for improving transparency and trust in Agentic AI systems. Existing Explainable AI (XAI) approaches are predominantly model-centric and fail to adequately capture the temporal and behavioral complexity of agentic systems. To bridge this gap, we propose Explainable Action Chains (XAC), a behavior-centric framework that represents agent decisions as temporally ordered units linking goals, states, actions, rationales, and outcomes. The framework is integrated into a modular Agentic AI architecture following the Goal–Plan–Act–Reflect cycle, enabling explanation generation as an inherent part of the decision process rather than a post-hoc addition. A prototype implementation is developed and evaluated through a comparative study between a baseline agent and an XAC-enabled agent under identical conditions. Experimental results demonstrate that XAC significantly increases perceived transparency and user trust without significantly affecting task success, while introducing only a modest increase in decision time. These findings validate the feasibility of behavior-level explainability for autonomous AI systems. The proposed framework contributes to the design of more transparent, accountable, and trustworthy Agentic AI systems, and provides a foundation for future research in explainable autonomous decision-making.