Enterprise networks are evolving from relatively static connectivity fabrics into distributed computing systems that span campuses, data centers, clouds, edge nodes, remote users, software-defined overlays, and programmable transport. This heterogeneity makes manual configuration and reactive operations increasingly inadequate. This review synthesizes recent research on artificial intelligence for network automation and develops an AI-powered intelligent network automation framework for future-ready enterprise computing systems. The review integrates evidence from machine learning for networking, intent-based networking, zero-touch management, network telemetry, network digital twins, programmable data planes, anomaly detection, federated learning, and explainable AI. The proposed framework organizes automation as a governed closed loop that translates business intent into machine-verifiable objectives, collects multi-layer telemetry, constructs contextual network state, applies predictive and decision models, validates actions in a digital-twin or policy sandbox, executes changes through software-defined and programmable infrastructure, and continuously verifies outcomes. The synthesis shows that effective autonomy depends less on a single algorithm than on the coordination of data quality, model lifecycle management, policy constraints, observability, security, and human oversight. Key research gaps concern cross-domain state representation, safe reinforcement learning, explainability, concept drift, trustworthy intent translation, multi-vendor interoperability, energy-aware optimization, and methods for measuring automation risk. The review contributes a practical architecture, an evaluation logic, and research directions for enterprises seeking higher resilience, agility, and computing efficiency without surrendering operational accountability.