<p>The rapid growth of connected smart devices and sensors has enabled intelligent systems development, such as autonomous vehicles, real-time health monitoring, and responsive network infrastructures. These applications are latency-sensitive and require consistent Quality of Service (QoS) guarantees. However, managing such systems is challenging due to fluctuating workloads and the need to meet service level agreements (SLAs), highlighting the need for dynamic resource management solutions at the edge areas that remain underexplored. To address these issues, this paper introduces a novel method for autonomous resource scaling and offloading in edge environments named JOARS (Joint Offloading and Auto-Scaling). JOAS employs meta-heuristic optimization to make concurrent decisions on when and where to offload tasks, and how to scale edge resources accordingly. The design prioritizes local offloading to nearby edge nodes to decrease latency and improve efficiency. An integrated edge intelligence module enables real-time autonomous decisions based on latency, cost, and task failure probability. Extensive experiments demonstrate that JOAS effectively balances system performance, cost, and QoS compliance under varying workload conditions. This work contributes to advancing scalable, resilient Internet of Things (IoT)-edge architectures capable of supporting future smart applications. According to the simulation results, the proposed approach facilitates a decrease in resource costs by 53% and response latency by 36.7%, while sustaining a similar level of task reliability compared to classic optimization methods in related research.</p>