Voice communication is insane? I suspect you are ignorant of what it is like to actually fly a large aircraft into a busy airport. Fault-tolerant and highly available hardware must facilitate low-latency, single-threaded communication with high semantic density in order to achieve multi-dimensional consensus in a safety-critical, heterogeneous, adversarial environment.
There is some interesting research that captures this sentiment and shows how complex a solution might need to be (replace "faulty agent" with "human error"): https://www.sciencedirect.com/science/article/abs/pii/S00051...