Running a mobile network has been rule-based automation: conditions are detected, predefined responses fire. Agentic AI proposes something different, systems given goals that decide for themselves how to reach them, and next-generation networks are an obvious candidate because their configuration space long ago outgrew what people can hand-tune.

The authors note that existing surveys cover the two fields separately, which leaves the connective tissue unexamined: how agents would integrate with network protocols, how their behaviour should be evaluated, and how any of it aligns with standardisation.

Those gaps are where the difficulty actually lives. Telecoms runs on interoperability agreements between vendors and operators, so an autonomous controller that cannot be specified in a standard, or whose behaviour cannot be evaluated against one, does not get deployed regardless of how well it performs in a lab.