Living Glossary

Every term in the series, explained for humans. Click any entry to explore.

10 terms

Directive

operations

A documented set of instructions for completing a specific task or process.

What it does

Tells the AI (or a human) exactly what to do, in what order, with what tools, and what to do when something goes wrong.

Without it

Knowledge lives in one person's head. When they're gone, so is the process.

Scary → Normal

"Write a directive with edge cases and escalation rules" → "Write down what you do, in order, including what to do if something goes wrong."

Introduced in Stage 1

CLAUDE.md

infrastructure

The instruction document that tells an AI how your organization works.

What it does

Contains your values, processes, guardrails, and decision-making rules. The AI reads it before every action.

Without it

Your AI is generic. It doesn't know your values, your style, or your boundaries.

Scary → Normal

"Write an AI instruction document" → "Write a letter to a new team member explaining how your organization works."

Introduced in Stage 2

Script

infrastructure

A small program that does one specific job reliably every time.

What it does

The deterministic execution layer — handles API calls, data processing, and file operations without AI judgment.

Without it

The AI tries to do everything itself. 90% accuracy per step = 59% success over 5 steps.

Scary → Normal

"Execute a Python script in the deterministic layer" → "Run a tool that does one job perfectly every time."

Introduced in Stage 2

3-Layer Architecture

infrastructure

The organizing principle: directives (what to do), orchestration (decision making), execution (doing the work).

What it does

Separates human intent from AI judgment from reliable tools. Each layer does what it's best at.

Without it

Everything is tangled together. Errors compound. Changes break unrelated things.

Scary → Normal

"Implement a 3-layer architecture" → "Separate your recipes, your chef, and your kitchen appliances."

Introduced in Stage 2

VOSPA

operations

Values → Organization → Story → People → Automation. A framework for seeing your entire organization through five concentric layers.

What it does

Diagnostic lens that reveals which layer needs attention first. Core outward, not linear.

Without it

You fix symptoms without seeing the layers underneath. Automate on top of broken values = scaled chaos.

Scary → Normal

"Apply the VOSPA diagnostic framework" → "Look at your organization through five lenses, starting from the core."

Introduced in Stage 1

Self-Annealing

operations

When something breaks, fix it and update the instructions so it never breaks that way again.

What it does

The feedback loop that makes your system antifragile — it gets stronger from every failure.

Without it

Same errors happen again and again. Fixes are forgotten. The system never improves.

Scary → Normal

"Implement a self-annealing feedback loop" → "When something breaks, write down the fix so it never breaks that way again."

Introduced in Stage 3

Hook

infrastructure

A rule that runs automatically to catch mistakes before they happen.

What it does

Deterministic guardrails — no AI judgment, just rules. If condition X, prevent action Y.

Without it

You rely on memory (human or AI) to catch every edge case. Memory fails.

Scary → Normal

"Deploy a deterministic pre-tool-use hook" → "Set up a rule that automatically catches mistakes before they happen."

Introduced in Stage 5

Vault

infrastructure

One place where everything your organization knows lives — structured, searchable, and yours.

What it does

A knowledge base in markdown files, organized by folders. Human-readable and machine-searchable.

Without it

Knowledge is scattered across email, docs, Slack, and memory. Finding anything takes minutes. AI can't access it.

Scary → Normal

"Set up a structured knowledge vault with markdown files" → "Create an organized folder of text files that you and your AI can both read."

Introduced in Stage 4

Memory System

infrastructure

How an AI remembers important things between conversations.

What it does

Persistent files indexed by topic. The AI reads them at the start of each session to recall context.

Without it

Every conversation starts from scratch. The AI forgets everything you told it yesterday.

Scary → Normal

"Create a persistent memory system with topic file indexing" → "Set up a notebook where your AI remembers important things between conversations."

Introduced in Stage 4

Agent

infrastructure

A specialized AI worker assigned to handle a specific type of task.

What it does

Like a team member with a job description — the research agent searches, the writer writes, the reviewer reviews.

Without it

One AI tries to do everything. Quality suffers. No specialization.

Scary → Normal

"Configure agent assignments with escalation boundaries" → "Give specific tasks to specific specialists, and tell them when to stop and ask you."

Introduced in Stage 3