Agentic AI for intelligent battery management systems: From diagnosis to autonomous control

Battery management systems (BMSs) are evolving from threshold-based supervisory units and isolated predictive modules toward intelligent, closed-loop decision architectures capable of diagnosis, prognosis, optimization, and autonomous control. This review examines how agentic artificial intelligence extends conventional AI-enabled BMS research by integrating state awareness, risk-informed reasoning, goal-driven planning, and safety-constrained action across cell-, pack-, and system-level operations. Rather than merely cataloguing prior studies, the review compares major groups of works spanning data-driven state estimation, degradation modeling, reinforcement learning-based control, hybrid physics-informed methods, digital twins, and emerging multi-agent strategies, highlighting their respective strengths, operational assumptions, scalability limits, interpretability, and deployment readiness. The analysis shows that prediction-oriented approaches remain dominant, while comparatively fewer studies achieve robust closed-loop autonomy under uncertainty, aging, thermal coupling, and heterogeneous usage conditions. To address this gap, the paper proposes the Agentic Battery Management Stack (A-BMS), a six-layer framework linking multimodal sensing, health inference, predictive risk assessment, decision intelligence, autonomous control execution, and human oversight. The review concludes by identifying key barriers and research priorities for trustworthy, adaptive, and certifiable autonomous battery management.