Large Language Models for Agentic NetOps and AIOps: Architectures, Evaluation, and Safety

The article discusses the integration of large language models in network operations (NetOps) and artificial intelligence for IT operations (AIOps), highlighting their role in tasks such as incident investigation and configuration synthesis. It emphasizes the importance of agent-based workflows that follow strict permissions and checks, advocating for a workflow-centered evaluation approach that includes measures like trace quality and sandboxed testing to ensure operational reliability. The findings underscore the necessity for a constrained approach to autonomy in these systems to mitigate security and governance risks, ensuring that outputs are reliable and auditable.