Modern supply chains operate under persistent uncertainty caused by fluctuating transportation costs, demand pressure, and weather-related disruptions. Traditional rule-based logistics strategies are often limited in their ability to adapt dynamically to such changing operating conditions. This study proposes an adaptive artificial intelligence-driven routing framework using reinforcement learning for real-time supply chain decision-making. The proposed system integrates real-time public economic indicators and weather signals into a simulated digital twin environment to model routing trade-offs among cost, delivery performance, and disruption exposure. Three decision policies were evaluated: a random baseline, a heuristic routing strategy, and a Proximal Policy Optimization (PPO) agent. Experimental results showed that both the heuristic and PPO policies significantly improved operational performance compared to the random baseline. While the heuristic and PPO policies achieved comparable on-time delivery rates of approximately 72–73%, the PPO agent demonstrated lower disruption exposure, indicating a more resilience-oriented routing strategy under uncertain environmental conditions, whereas the heuristic policy maintained slightly better cost efficiency. These findings demonstrate the potential of reinforcement learning to support adaptive and resilient logistics decision-making under dynamic external conditions. The primary contribution of this study lies in integrating real-time economic indicators, weather-driven disruption signals, and reinforcement learning-based routing optimization within a unified digital twin supply chain environment. Although the framework was evaluated within a simulated environment, the results provide a practical foundation for future enterprise-scale adaptive logistics systems.