Warehouses increasingly depend on the interaction between physical stock movements and digital decisions. Yet robotic picking, inventory counting, demand prediction and replenishment are frequently evaluated as separate applications, obscuring the consequences of errors passed between them. This critical narrative review examines how artificial intelligence can connect these functions in Saudi Arabia's developing logistics network. Thirty DOI-identified publications from 2020–2025 were mapped across robotic fulfilment, inventory perception, forecasting, sequential inventory control and implementation governance. The synthesis differentiates findings demonstrated in operational settings from optimisation or simulation results, and proposes an integrated decision architecture linking verified stock states to probabilistic demand and constrained ordering. Evidence indicates that robotics can reduce travel and counting effort, while machine-learning forecasting and adaptive control offer decision support under variable demand; however, benefits cannot be assumed to transfer across warehouse layouts, product classes or service conditions. Inventory record integrity emerges as a prerequisite for dependable forecasting and automated orders, rather than a secondary outcome of automation. For Saudi organisations, deployment priorities include multilingual master-data governance, extreme-temperature resilience, interoperability, supplier lead-time visibility and workforce safety. The review proposes staged validation measures that combine record accuracy, fulfilment quality, stockout exposure, intervention rates and energy use. It concludes that the most defensible pathway is auditable, human-supervised automation in which perception, prediction and procurement are jointly evaluated.