That’s Not a Job for an LLM: The Right Way to Apply AI to Network Operations (Sponsored)

Why It Matters

Understanding the true strengths and limits of LLMs versus traditional AI ensures network teams deploy the right technology, improving reliability while avoiding costly mis‑automation.

Key Takeaways

• LLMs are not a replacement for traditional ML in network ops.
• Fuzzy logic already powers capacity planning and DDoS detection.
• LLMs excel at language tasks, but struggle with precise mathematical calculations.
• Telemetry data is best processed with statistical ML, not tokenized LLM inputs.
• Trust‑but‑verify remains essential when deploying LLM‑driven automation in networks.

Summary

The Heavy Networking episode cuts through AI hype to explain how different artificial‑intelligence techniques actually affect network operations. Host Ethan Banks and guest Avi Freriedman, founder of Kent, argue that large language models (LLMs) are only one part of a broader AI toolbox that includes decades‑old expert systems, fuzzy logic, and statistical machine learning.

They trace AI’s networking roots back to rule‑based expert systems and early fuzzy‑logic controllers used for capacity planning and DDoS detection. Modern ML models ingest massive telemetry streams, applying pattern‑matching and statistical inference to predict outages or attacks. By contrast, LLMs rely on massive text corpora, generating predictions about the next token rather than solving precise numeric problems.

Avi likens LLM evolution to GPS: early versions were useful but error‑prone, while today they’re increasingly reliable but still require human oversight. He cites Kent’s production DDoS mitigation, built on ML, as a concrete success, and warns that LLMs can hallucinate or produce inconsistent answers, especially when asked to perform calculations or enforce deterministic network policies.

The takeaway for operators is clear: match the tool to the task. Use statistical ML for high‑volume telemetry analysis, retain fuzzy‑logic rules for deterministic decision‑making, and reserve LLMs for language‑heavy workflows such as documentation or troubleshooting guidance—always with a "trust‑but‑verify" mindset.

Original Description

On today’s sponsored Heavy Networking, we get off the AI hype train to talk about how different artificial intelligence techniques usefully impact network operations---and where they aren't a fit. The various forms of AI represent a set of tools that, like any tool, have use cases, capabilities, and limitations. Our guest is Avi Freedman, CEO and founder of Kentik. Avi is here to share his AI knowledge so that you know what’s real, not hype, and how AI will realistically impact network operations.

Links:

kentik.com/packetpushers

AI Advisor - https://www.kentik.com/product/ai-advisor/

Practical Guide to Modern Network Telemetry by Avi Freedman and Leon Adato:

Oreilly Library - https://www.oreilly.com/library/view/practical-guide-to/9798341608917/

Kentik Ebook - https://www.kentik.com/go/ebook/oreilly-guide-to-modern-networking-telemetry/

Avi Freedman on LinkedIn - https://www.linkedin.com/in/avifreedman/

Heavy Networking is the flagship show of the Packet Pushers network. Visit our website to find more great networking and technology podcasts, along with tutorial videos, the Human Infrastructure newsletter, and loads more resources for building your IT career. https://packetpushers.net