Abstract: Artificial intelligence is reshaping modern communication networks, bringing a critical dilemma regarding future network development. One conservative opinion supports incremental evolution, which retains the classic layered network architecture to continuously optimize key metrics such as bandwidth and latency. In this paradigm, AI acts merely as an auxiliary tool, ensuring mature ecosystems and low renovation costs for practical deployment. In contrast, another perspective advocates fundamental network restructuring. Traditional layered structures exhibit obvious limitations in AI-native scenarios, including vehicular networks, intelligent healthcare, and large-model services. With powerful learning capabilities, AI can break rigid network layers, rebuild communication protocols, and further optimize resource allocation for diverse intelligent services. This panel brings together experts from multiple research backgrounds. Panelists will debate the two technical routes from both industrial and academic perspectives. This discussion aims to clarify the advantages and limitations of evolution and restructuring, and explore reasonable technical directions for next-generation AI-driven networks.