AI is a demand story for networks as well as a technology story. Real-time voice agents, inference calls and AI-enabled devices are less tolerant of delay, jitter and packet loss than human users, and they push traffic upstream, while mobile networks were built for downlink-heavy video. This session looks at what real-time AI workloads need from the network end to end, how measures such as upload latency under load and upstream jitter are joining download speed as tests of performance, and how the location of inference, whether on the device, at the edge or in a regional cloud, changes what users experience. It then turns to AI-RAN, the integration of AI processing into the radio access network, and how it can improve spectral efficiency, energy use and performance while creating new revenue opportunities. Panelists will discuss uplink/downlink rebalancing, new equipment and edge compute at cell sites, and the implications for power, space and backhaul. Attendees will leave with a practical checklist for judging network readiness for AI workloads and a clear view of how AI is reshaping both RAN design and the infrastructure that supports it.