Conventional Artificial Neural Networks (ANNs) present challenges for edge devices due to their high computational and power demands. Spiking Neural Networks (SNNs) offer a power-efficient alternative by transmitting information via sparse binary spikes, significantly reducing energy consumption and latency. However, SNNs often face limitations in task accuracy compared to ANNs. This paper explores hybrid SNN-ANN architectures as a promising solution to reconcile the trade-off between efficiency and accuracy. We survey the neuro-computational foundations of SNNs, including spike encoding and learning strategies pertinent to edge deployment. Furthermore, this work articulates the rationale behind hybrid SNN-ANN designs, analyzing how multi-timescale architectures and ANN-to-SNN migration strategies address the efficiency-accuracy dilemma. Finally, we synthesize the current state of hybrid models in wireless edge intelligence, highlighting their capacity to achieve substantial power reductions with minimal impact on accuracy, thereby enabling robust, low-latency inference in resource-constrained environments.