This is where I keep my AI research, the frameworks I'm working out, and the explainers I write to teach this stuff. I come at all of it from the network side. After 20+ years of keeping packets moving, I can't look at a model, an agent or the plumbing behind them without asking the same questions I'd ask about any network: what are the layers, what depends on what, and where's it gonna break at 2 AM?
Some of what's here is a working draft I'm thinking through out loud, some of it's plain-English teaching material, and some of it's code you can run on your own hardware. When something's a draft or an opinion, I'll say so. Every card links to the full write-up, so start wherever you like.
Start here
New to this, or teaching somebody who is? This is the order I’d read things in. Each one builds on the one before it, and it ends at the framework that ties them together.
- How AI really worksTokens, tensors and transformers, and why the hardware mattered as much as the ideas.
- What happens when you hit EnterOne prompt traced like a packet, from DNS and TLS to prefill, decode and the token stream.
- Why AI is a networking problemThousands of GPUs only act like one computer if the network between them never blinks.
- Run your own AI at homeVRAM, quantization, two sample builds and a 15-minute Ollama quick start.
- Agents and network engineeringWhat agents can actually do today, starting with a blank router in my home lab.
- The AIM modelPut it all together with a seven-layer way to think about agents.
Frameworks & Mental Models
Network engineers have OSI. When something breaks, we know which layer to look at first. These are my attempts to give AI agents the same kind of map.
My seven-layer way of thinking about AI agents, built the way OSI taught us to think about networks. MCP sits at Layer 3 as the narrow waist, Skills live at Layer 4, and Trust and Telemetry run down the side of every layer. It's a working draft and a mental model, not a standard.
Remember the layers top down, 7 to 1
Agent Infrastructure & MCP
The plumbing under the agents: how they reach tools and data, what it costs to run the models, and where the trust boundaries go.
Framework section
Where MCP sits, and why
Why I put the Model Context Protocol at Layer 3 as the narrow waist of the agent stack, and what changed in the 2026-07-28 spec that made me comfortable doing it.
Framework section
Which layer is broken?
A symptom-to-layer troubleshooting table for agents, run the same way you'd chase a network outage: bottom up, top down, or split it at the middle.
Guide
Run Your Own AI at Home for the Price of a Gaming PC
VRAM, quantization, two sample builds and a 15-minute Ollama quick start. The cheapest way I know to get a local model running.
Free tool
AI Rent vs. Own vs. Hybrid: A TCO Model in Excel
A free Excel model for deciding whether to rent frontier AI through APIs, own the hardware, or go hybrid. It counts power, colo, depreciation, staffing and utilization, not just token burn.
AI for Network Operations
Where agents and automation meet real routers and switches, and the guardrails I'd want before letting them anywhere near production.
Field notes
AI Agents and Network Engineering: The Train Is Coming, So Get On It
Two agents in my home lab took a blank Cisco router to a working config and a port map, with a QA agent checking the first one's work. Plus why enterprise and OT networks aren't ready to hand over the keys yet.
Guide
Netmiko vs NAPALM vs Nornir
The Python automation tools an agent workflow ends up leaning on, and why compare-before-commit matters even more when an AI drafted the code.
Build log
Agentic AI on Patch Duty
An AI agent handles the OS and app updates across the three Raspberry Pis in my family streaming stack, so patching happens on schedule instead of whenever I remember.
Open Source & Experiments
Code you can clone and run on a machine you control.
Open source
StreamSwarm
A distributed network monitoring platform. A central Flask server schedules tests across clients and charts the results, with machine learning diagnostics (Scikit-learn) that run locally, plus PDF reporting. MIT licensed.
Everything else
All my projects
Zixt, NerdGram1337 and the rest of what I publish live on the Software page and on GitHub at NetworkNerd1337.
Writing: AI, the Industry and the Hype
Longer pieces on where AI is headed, the money behind it, and what's real versus what's marketing.
Analysis
AI Inference Rationing: Is This the Day the Music Died?
What happens when throttling heavy users becomes normal. Four scenarios with my odds, and five predictions for wide-scale adoption.
Analysis
Guppy-Level Death: The Neocloud Warning
What a neocloud is, how GPU-backed debt works, and why it rhymes with the 2000 telecom bust and the 2008 housing crisis.
Explainer
AI Reverse Engineering Isn't Magic
What actually happens when software gets "cloned" with AI, from Ghidra to open-weight models to a whole lot of testing.
Explainer
Post-Quantum Cryptography, AI and Q-Day
What quantum actually breaks, what AI can and can't do to the math, and why the migration clock is running anyway.
Looking for everything? Browse every AI / LLM post on Signal & Noise.
Questions and corrections
Got a correction, a question, or think something here is flat wrong? I’d rather hear it than not. Send me a note through the contact page, or find me as @NetworkNerd1337 on LinkedIn and Mastodon.