Agentic AI for Autonomous Cloud Infrastructure Management: A Multi-Agent Framework for Fault Detection, Root-Cause Analysis, and Self-Healing

Abstract

The increasing complexity of cloud-native infrastructure, characterized by distributed microservices, dynamic workloads, and heterogeneous computing environments, has exposed limitations in conventional fault-management systems. Existing approaches frequently suffer from fragmented fault detection, inaccurate root-cause localization, delayed incident response, and limited autonomous recovery capabilities, compromising infrastructure reliability and operational efficiency. This study proposes an Agentic AI-based Multi-Agent Autonomous Cloud Management Framework (AIC-MAC) to enhance intelligent fault detection, root-cause analysis, and autonomous self-healing in Kubernetes-based cloud environments. The framework integrates six specialized agents responsible for infrastructure monitoring, anomaly detection, root-cause diagnosis, remediation planning, policy enforcement, and recovery verification. It incorporates multivariate telemetry analysis, dependency-aware causal reasoning, evidence-grounded decision-making, and risk-constrained remediation. A controlled experimental methodology is designed using Kubernetes-hosted microservices, observability tools, and fault-injection scenarios. The proposed evaluation assesses fault detection precision, recall, F1-score, root-cause localization accuracy, mean time to detect, mean time to recover, and remediation success rates against conventional monitoring and AIOps baselines. Experimental findings remain pending implementation and validation. AIC-MAC establishes a structured foundation for coordinated, explainable, and safety-aware cloud infrastructure management, with the potential to improve operational resilience and autonomous fault recovery.