AI is transforming supply chain from reactive to predictive and autonomous. Demand sensing (short-term forecasting): supplement historical sales data with external signals - weather forecasts (ice cream demand), Google Trends (product interest), social media sentiment, competitor pricing. Gradient boosting with these features achieves 15-25% better short-term accuracy vs pure historical models. Demand signal timing: some signals lead demand by hours (social), others by days (weather), model must capture these lag relationships. Supplier risk prediction: train classifier on suppliers with features: financial health metrics (Dun & Bradstreet), news sentiment (NLP on company news), geographic risk scores, historical performance. Predict probability of disruption 90 days ahead. Integrate with procurement to flag at-risk suppliers for dual-sourcing. Disruption detection: real-time monitoring of supplier news, port congestion data, commodity prices, geopolitical events using NLP. Alert supply chain teams to potential disruptions before they materialize. Autonomous procurement: RL agent for routine purchase order generation based on inventory levels, lead times, and forecasts. Requires guardrails: approval workflow for orders above threshold, anomaly detection to flag unusual patterns. Implementation stack: Databricks for data engineering + ML training, Feast feature store for real-time feature serving, FastAPI for inference, Grafana for monitoring.