Early detection and intelligent management of plant diseases are essential to ensure food security and farm profitability. We present an integrated AI framework combining deep convolutional neural networks (CNNs) for early banana leaf disease diagnosis and a reinforcement learning (RL) agent for precision farming decisions (pesticide dosing, irrigation, nutrient scheduling). The CNN uses transfer learning and extensive augmentation to achieve robust recognition across diverse agro-climatic conditions. The RL agent formulates actions as a Markov Decision Process (MDP), balancing disease suppression, resource cost and environmental impact. We evaluate the framework on a diverse dataset of 15,000 annotated banana leaf images collected across three continents, and on simulated field environments. The system attains 96.8% classification accuracy and demonstrates substantial reductions in chemical usage (˜25–35%) in simulated deployments while maintaining or improving estimated yield. We also present interpretability using Grad-CAM and SHAP and discuss practical deployment via IoT and edge devices.