📦 Fundamentals & Glossary
The AIOS mental model and vocabulary. What an AI operating system is, its 6 layers, and each term defined inline. No one moves forward without understanding the terms.
How to read: one domain (on the left) is "engineered" across the 6 layers (center, in cyan) that together form a living OS (on the right). This learning path covers the first three: the concept, the layers, and the glossary.
Learning path map
Detailed content
📦 What is an AI Operating System
Command center for a domain: the concept, the origin, the examples, and the anti-pattern “install instead of cultivate.”
An AIOS is a command center for a specific domain of your life or business—not a single tool.
It's the foundation of everything: without defining the concept, you'll confuse an OS with "just another app."
Domain · command center · tailored to each area of life.
The concept was born from workshops where the author sought the visual analogy that clears up confusion about AIOS.
Understanding its origins shows that it’s more art than science—and that a simple mental model is enough.
Simple mental model · visual analogy · art > science.
Five real OSs: Freedom (residencies/passports), Tax (laws/auditing), Skool (ads), Health, Marketing.
The examples make “one OS per domain” concrete and give you a reference for your own.
One OS per domain · each with a different composition.
In practice, an OS is a handful of folders + an identity file (CLAUDE.md).
Demystifies it: you don't need heavy infrastructure to get started, just a file structure.
Folders · master file · plumbing > model.
The concept works with any harness: Claude Code, Codex, open code. The folder changes, not the idea.
You aren’t tied to one tool; your investment survives changes.
Tool-agnostic · portable · same idea, different folders.
There's no fixed formula: each OS is designed by feel, then adjusted through use.
Frees you from the search for the “perfect setup” — what matters is iterating.
Design > engineering · iteration · judgment.
Treating the OS as a product ready to install is the mistake. It's a garden you cultivate.
Set expectations: 80% in a weekend, solid only with daily use and auditing.
Cultivate > install · living system · trial by fire.
🍕 The 6 layers and the pie chart
Why the pie isn’t split evenly, the order that works, and the #1 mistake: jumping to agents too soon.
Identity, Substrate/Context, Rules & Hooks, Skills, Tools, Agents.
It’s the structural vocabulary of every OS — a reference for the technical learning paths.
6 layers · foundation vs. action.
Slices vary by domain: in a Tax OS, context weighs ~34%; in others, skills or tools dominate.
You stop treating all layers as equal and invest where it hurts most.
Distribution by domain · unequal percentages.
The build sequence: Identity → Substrate → Rules → Skills → Tools → Agents.
Building out of order wastes effort; the foundation supports capability.
Order of operations · foundation before capability.
Most people start with agents—the last layer—before they have a foundation.
Avoids the most common and costly mistake beginners make.
Agents last · prerequisites first.
Static layers (Identity, Substrate, Rules) support the dynamic ones (Skills, Tools, Agents).
Organizes the course: Track 2 = foundation, Track 3 = action.
Static = back of house · dynamic = front.
Preview of how Tax, Sales, Support, Content, and Consulting redistribute the layers.
Previews Track 5 and practices asking “which layer matters here?”
Tax ~34% context · Sales ~30% · preview.
In a new domain, it’s sometimes worth gathering context before writing the identity.
The order is a guide, not dogma: you adjust it based on what you know (or don't know).
Naive in the domain · context-first.
📖 Essential Glossary
Foundation: define each term inline. Identity, substrate, rule vs. hook, skill, agent, CLI/MCP/API, nugget, context rot, reverse meta-prompting, raw vs. synthesized.
The soul of the OS: who it is, whom it serves, what it refuses. It lives in the CLAUDE.md.
It’s the first file the OS reads — it sets the tone for everything.
Soul file · concise · < 30 lines.
Substrate = stored reference material; context = what goes into the conversation when needed.
Distinguishing the two prevents bloating the context window.
Reference · injected on demand.
Rule = “do not enter” sign (strong suggestion); hook = locked door (deterministic).
You decide what can be "flexible" and what must be unbreakable (PII, money).
Nondeterministic vs. deterministic.
A skill is a recipe card: a packaged task that runs the same way every time.
It's the unit of capability — starts manual, then becomes a recipe.
Verb · repeatable · 1 task.
The chef who chooses the tool and tastes the dish: orchestrates several skills with judgment.
Separate the "verb" (skill) from the "worker" (agent)—only promote routines you already do by hand.
Judgment · orchestrates skills · review gate.
CLI = a lean remote control; MCP = a power adapter; API = a data port; skill = a recipe.
Knowing how to choose avoids unnecessary infrastructure (all infrastructure becomes maintenance).
4 connectors · MCP loses relevance · read-only CLI.
Nugget = super-distilled summary; context rot = context that grows stale; reverse meta-prompting = asking AI to distill the conversation into a skill.
These are the concepts most people ignore — and what separates a living OS from a dead one.
Distill · decay · recipe after cooking.
Folder raw/ (raw, kept as a backup) versus a distilled folder (nuggets that go into the context).
It’s the data pattern that repeats in every domain: pantry vs. finished dish.
raw > synthesized · inject only nuggets.