In today’s interconnected environments, the rapid growth of Internet of Things (IoT) devices has led to a dramatic rise in data traffic, placing increasing demands on energy-constrained networks. Traditional routing protocols are often unable to meet the challenges posed by battery-powered IoT devices, which must balance efficient data transmission with limited energy availability. To address these challenges, we introduce ARROW (Additive Reinforcement-driven Routing for Optimal Wireless networks), a new paradigm in IoT routing that combines reinforcement learning with additive encoding embedded directly in packet headers. This integration enables lightweight, decentralized learning, allowing each node to dynamically adapt its routing behavior based on real-time network feedback and in-packet energy metrics. In this paper, we define ARROW-IoT as a specific implementation of the ARROW paradigm, applied to IoT networks. It leverages Q-learning and additive header encoding to enable energy-aware, decentralized routing decisions in dynamic, resource-constrained environments. ARROW highlights how in-packet intelligence can support autonomous, energy-aware decision-making across dynamic networks with minimal overhead.