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

LLMs have sucked all the oxygen out of the AI conversation — but AI is much more than just LLMs, and network engineers have been using AI techniques (machine learning, statistics, fuzzy logic, expert systems, neural networks) for decades. So what should LLMs be doing in network operations, what shouldn't they be doing, and how do agentic AI architectures fit in? In this Heavy Networking episode (sponsored by Kentik), Packet Pushers hosts Ethan Banks and Drew Conry-Murray sit down with Avi Freedman, founder of Kentik, for a candid, jargon-aware conversation about the right way to apply AI to network operations. Avi has seen AI from both sides — using it to make Kentik's product suite more useful, and watching the industry overpromise on "AI ops" for years before LLMs ever arrived. What you'll learn: Why AI is much broader than LLMs (ML, fuzzy logic, expert systems, neural networks) Why anything math-heavy is still the domain of ML and statistics, not LLMs Why LLMs hallucinate and what guardrails look like How Kentik AI Advisor uses LLMs to evaluate other LLMs to catch hallucinations Why agentic AI doesn't have to be LLM-based Where to draw the line between read and write tasks for autonomous AI Why blast radius matters more than likelihood of success for AI-driven automation Why explainable AI (XAI) and tunable determinism are likely the next major directions Featuring Avi Freedman (Founder, Kentik), Ethan Banks, and Drew Conry-Murray (Packet Pushers). ─── RELATED RESOURCES ─── More Kentik content on AI: AI NetOps: How AI and Machine Learning Transform Network Operations:    • AI NetOps: How AI and Machine Learning Tra...   How AI Transforms Network Chaos to Clarity:    • AI for Network Operations: From Traffic Sp...   3 Service Providers on Kentik AI Advisor:    • Video   Agentic Network Operations: ServiceNow + Kentik:    • Video   Streaming Telemetry and SNMP Monitoring with Kentik NMS:    • Streaming Telemetry and SNMP Monitoring wi...   Learn more on kentik.com: Kentik AI: https://www.kentik.com/solutions/kent... Kentik AI Advisor: https://www.kentik.com/product/ai-adv... Kentik for Packet Pushers: https://www.kentik.com/go/packet-push... Best Network Monitoring Tools for 2026: https://www.kentik.com/kentipedia/bes... Kentipedia: https://www.kentik.com/kentipedia/ ─── ABOUT KENTIK ─── Kentik is the network intelligence platform for modern infrastructure teams. Kentik provides network observability, DDoS protection, cloud cost optimization, BGP monitoring, and AI-powered network analytics — including Kentik AI Advisor — for enterprise, cloud, and service provider networks. Start a free trial: https://www.kentik.com/go/get-started/ Subscribe:    / @kentikhq   #Kentik #AI #NetOps #LLM #AgenticAI #NetworkAutomation #MachineLearning #NetworkIntelligence #PacketPushers CHAPTERS: 00:00:00 Getting Off the AI Hype Train: Why LLMs Aren't All of AI 00:01:39 Meet Avi Freedman, Founder of Kentik 00:03:09 The Pre-LLM History of AI in Networking 00:04:03 Machine Learning, Statistics, Fuzzy Logic, and Expert Systems in NetOps 00:09:51 Why LLMs Are Additive to Older AI Techniques, Not a Replacement 00:11:13 When to Use ML vs LLMs (Hint: Anything Math-Heavy) 00:12:02 What a Large Language Model Actually Is (and Isn't) 00:13:55 Why LLMs Hallucinate: No Semantic Understanding, Just Probability 00:14:30 What "Reasoning Mode" Really Does in Models Like Gemini and Claude 00:17:10 Hallucination, Edge Conditions, and Why LLMs Don't Know What They Don't Know 00:19:57 Determinism vs Non-Determinism in LLMs for Network Operations 00:20:35 How Kentik AI Advisor Adds Guardrails to LLMs 00:24:48 What Is Agentic AI? (And Why It Doesn't Have to Be LLM-Based) 00:25:09 How Agents, Tools, and MCP Actually Fit Together 00:30:01 How Kentik Built Its Internal Agentic Architecture for AI Advisor 00:30:42 ServiceNow Integration: Cross-Platform Agentic Troubleshooting 00:32:01 When Should You Build Your Own AI vs Use What Vendors Provide? 00:35:53 What Kinds of Questions Can You Ask an LLM in Network Operations? 00:38:40 Why "Tribal Knowledge" Encoded as Prompts Beats Foundation Model Defaults 00:40:40 Information-Gathering vs Task-Performance: Where Are We? 00:43:09 Why LLMs Struggle with Network Architecture Planning Tasks 00:44:14 Augmented Intelligence: AI as a Force Multiplier 00:49:55 Router/Switch Code Quality: Why You'll Keep Finding Bugs 00:51:24 Where to Draw the Line Between Read and Write Tasks for AI 00:53:43 Blast Radius: A Better Planning Concept Than "Likelihood of Success" 00:55:43 Are LLMs at a Dead End? (No.) 00:56:30 Where AI Is Heading: Explainable AI and Tunable Determinism 00:58:06 What Is Kentik AI Advisor? 00:58:53 The O'Reilly Book on Network Telemetry 00:59:23 Closing Remarks