The integration of machine learning (ML) into healthcare offers significant opportunities to reduce prescription errors and support intelligent decision-making. This study investigates the development of an agentic AI framework for drug prescription using the publicly available Drugs200 dataset, which includes patient attributes such as age, sex, blood pressure, cholesterol, and sodium-to-potassium ratio. Three ML algorithms, Naïve Bayes, Decision Tree, and Random Forest, were implemented to classify patients into one of five drug categories. Results indicate a clear performance hierarchy across the models. Naïve Bayes achieved stable but modest accuracy, with testing performance ranging from 86% to 91%, though precision was limited for minority drug classes. Decision tree improved interpretability and achieved testing accuracies between 82% and 93%, but its performance fluctuated depending on class imbalance and data splits. Random Forest consistently outperformed both, delivering perfect results across all evaluation metrics, including 100% accuracy, precision, recall, and F1-scores, with zero misclassifications across all split ratios. The study demonstrates the feasibility of ML-based drug classification as a foundation for agentic AI in healthcare decision support.