Background: Autonomous artificial intelligence (AI) has the potential to transform decision-intensive environments; however, retail supply chains continue to rely on static forecasting models and manual inventory policies that degrade under volatility and demand drift. Current systems lack continuous learning, adaptive optimization, and integrated feedback loops required for real-time operation. Aim: This study aims to develop and evaluate a fully autonomous AI architecture that unifies demand forecasting, reinforcement learning (RL)–based inventory control, and automated MLOps-driven monitoring to achieve scalable, self optimizing retail supply chain performance. Method: A multi-model forecasting engine incorporating statistical, machine-learning, and deep-learning models—led by a Transformer-based architecture—is combined with an RL agent formulated as a cost-sensitive Markov decision process. Automated drift detection and retraining pipelines maintain continuous adaptation. Experiments use multi-year retail datasets and stress-test scenarios. Results: The autonomous architecture reduces forecasting error by up to 34 percent, lowers total inventory cost by more than 30 percent, and increases service-level performance relative to EOQ, (s, S), and forecasting-only baselines. Stress tests confirm resilience under demand shocks, supply delays, and seasonal reversals. Conclusion: Findings demonstrate that closed-loop AI systems can autonomously learn, adapt, and optimize retail operations, offering a scalable pathway toward self-optimizing supply chains.