This paper presents an advanced Edge AI framework that synergistically optimizes smart farming operations and microgrid energy management in resource-constrained environments. Our integrated solution addresses three critical challenges: Firstly, real-time crop health monitoring through a novel quantized convolutional neural network (Q-CNN) achieving 94.1% accuracy with only 4.2 MB memory footprint. While secondly, adaptive irrigation control demonstrating 30% water use reduction while maintaining crop yield. Finally, decentralized grid optimization via reinforcement learning that reduces power transmission losses by 22.5%. The system architecture combines lightweight machine learning models deployed on edge devices with a hierarchical decision-making framework, enabling sub-second response times for critical operations. Field validation using soybean farms and microgrid testbeds confirms the framework’s practical viability, showing 18% faster inference than cloud-based alternatives and robust performance under dynamic load conditions. Key technological innovations include mixed-precision quantization for efficient model deployment, attention mechanisms for improved feature extraction, and a hybrid reward function for grid stability. This research contributes to sustainable agriculture by providing a scalable solution that simultaneously improves water use efficiency (WUE) by 35%, reduces energy losses, and maintains sub-150ms fault recovery times. The complete system implementation, including hardware specification and model architectures, is presented with comprehensive performance benchmarks against state-of-the-art approaches.