With continuous advancements in the perception, generation, and reasoning capabilities of Large AI Models (LAMs), communication networks are rapidly transitioning into a new era characterized by intelligent connectivity and ubiquitous interconnection. This report systematically examines the integration pathways of LAMs and Agentic AI in key communication scenarios such as semantic communication and edge intelligence, and reveals the architectural evolution from data-driven to model-driven and ultimately to agent-driven paradigms. The report first introduces the design of the multimodal foundation model CommGPT, which is tailored for communication tasks. It elaborates on the construction of domain-specific datasets, the development of internal learning mechanisms, and the implementation of retrieval-augmented strategies. Based on this foundation, an Agentic AI system framework named CommLLM is proposed for 6G communication scenarios. This framework incorporates core modules including communication knowledge retrieval, collaborative task planning, and reflective evaluation. Subsequently, the report explores representative applications of LAMs in semantic communication, with a particular focus on the semantic communication foundation model M4SC. This model demonstrates strong potential in supporting multi-task, multimodal, and multi-scenario communication systems. Finally, with an eye toward future 6G networks, the report provides an in-depth analysis of the key challenges and opportunities associated with building autonomous, efficient, and evolvable intelligent communication systems empowered by LAMs and Agentic AI.