<p>Smart agricultural IoT systems face dynamic resource allocation challenges from device heterogeneity and environmental variability. In this paper, we propose a novel framework that combines 6G edge computing with reinforcement learning to optimise resource management. The framework exploits the ultra-low latency and high bandwidth of 6G for real-time communication and efficient task offloading, while reinforcement learning adapts the resource allocation strategy based on changing environment states and task requirements. A reward function is designed to balance latency reduction, resource utilisation and energy efficiency. Experimental results demonstrate 35% latency reduction and 25% resource utilization improvement over traditional approaches, with scalability validated under increasing device density and task complexity. This work provides a scalable and adaptive solution for smart agriculture, which can be used in areas such as precision irrigation, autonomous agriculture, and drone surveillance.</p>