5G edge network slicing faces a dual challenge: it must understand high-level operator intent while also reallocating scarce computing and bandwidth resources under dynamic contention. These requirements are coupled, making orchestration particularly difficult when conditions change rapidly.

The paper describes a conflict-aware autonomous orchestration approach that combines multi-agent systems with large language models. This design is contrasted with centralized optimization, which the abstract suggests can encounter problems, though the full discussion is not visible in the excerpt.

By leveraging LLMs, the framework aims to interpret operator intent more flexibly, while multi-agent systems handle distributed resource decisions. The approach is positioned as a way to reconcile the two coupled requirements in real time, potentially improving responsiveness in 5G edge environments.