The Everything Feed - All Packet Pushers Pods HN801: Will a Natural Language Interface (NLI) Replace Your CLI?
Oct 17, 2025
Scott Robon, host of the Total Network Operations podcast and a network automation expert, dives into the potential of natural language interfaces (NLIs) replacing command-line interfaces (CLIs). He discusses real-world vendor developments in NLIs and the challenges of LLM variability versus deterministic outputs. The conversation shifts to the implications of running AI infrastructure and the crucial role of data engineering. Robon emphasizes the ongoing need for engineers to understand automation tools while assuring that current AI is about augmentation, not replacement.
AI Snips
Chapters
Transcript
Episode notes
LLM As A Translation Layer
- The LLM often acts as a translator sitting atop many specialized models that do the heavy data work.
- Those specialist models can use non-LLM techniques (graphs, regression) better suited to specific network tasks.
Require Human Review For Changes
- Keep humans in the loop and surface the translated CLI or plan for review before applying changes.
- Expose the underlying CLI output so experienced engineers can validate intent and catch errors.
Real-Time Data Is The Hard Problem
- Real-time, high-volume telemetry and low-latency inference are core challenges for network NLIs.
- The compute and power costs to ingest continuous network data are substantial and ongoing.