Stage 01 · How models work · 7 frames
What an AI Model Actually Is
A file of numbers that ranks what comes next, and a lot of software built around it. Seven frames on which behaviour comes from which.
See the idea
Storyboards
Illustrated explainers of how applied AI systems work. Each one tells a single Systems article as a short visual story, each drawn around its own idea. Flip through it here, keep it as a PDF, and follow it back to the article for the whole argument.
StoryboardWhat an AI Model Actually Is
StoryboardRetrieval-Augmented Generation in Plain Terms
StoryboardPrompting Is Not the Skill You Think It Is
StoryboardFrom Agent Intent to Governed Execution
StoryboardHuman-in-the-Loop Is a System Design Choice
StoryboardWhat Counts as Good, and How to Check It
StoryboardWhat Has to Survive for Work to Continue
StoryboardWhen Is an AI System Ready for Real Use?
Stage 01What an AI Model Actually Is
Stage 01From Text to Tokens
Stage 01Training, Fine-Tuning, and Inference
Stage 02Retrieval-Augmented Generation in Plain Terms
Stage 02What Are Embeddings?
Stage 02Context Windows: What the Model Can See in One Run
Stage 02Why OCR Quietly Breaks Document AI
Stage 02Semantic Caching: Similar Is Not Safe to Reuse
Stage 03Prompting Is Not the Skill You Think It Is
Stage 03Structured Output and Why It Matters
Stage 03What a System Prompt Actually Is
Stage 03Prompt, Skill, Workflow, Agent: Telling the Layers Apart
Stage 03What Are Agent Skills, and What Makes One Reusable?
Stage 04From Agent Intent to Governed Execution
Stage 04Tool Use: When Language Triggers Actions
Stage 04AI Agents vs AI Workflows
Stage 04What Is an Agent Harness?
Stage 04What Is MCP? The Model Context Protocol in Plain Terms
Stage 04Policy-Governed MCP Runtimes: Trust Boundaries for Tool Execution
Stage 04Why the Loop Outlives the Model
Stage 05Human-in-the-Loop Is a System Design Choice
Stage 05When Chat Renders an Interface
Stage 05What Is Bounded Autonomy?
Stage 06What Counts as Good, and How to Check It
Stage 06What to Record So You Can Explain What Happened
Stage 07What Has to Survive for Work to Continue
Stage 07The I-7 Loop: A Working Method for Delegating to AI
Stage 07The Instruction Spec: Defining an AI-Assisted Software Change
Stage 08When Is an AI System Ready for Real Use?
Stage 08What Does Forward Deployed Mean?
Stage 08What Is a Forward Deployed Consultant (FDC)?
Stage 08Why Do AI Pilots Stall Before Production?
Across the mapHow the Parts of an AI Application Depend on Each Other
Across the mapWhat Systems Thinking Helps You Inspect
Across the mapWhat a Publisher Actually Controls- Stage 02 · Context & retrieval · 8 frames
Retrieval-Augmented Generation in Plain Terms
Finding the right document and handing it to the model are two different steps. Eight frames on where the evidence gets lost.
- Stage 03 · Instructions & contracts · 7 frames
Prompting Is Not the Skill You Think It Is
A prompt shapes the odds; it doesn't enforce anything. Seven frames on asking versus checking, and the skill that actually matters.
- Stage 04 · Action & authorization · 11 frames
From Agent Intent to Governed Execution
What decides whether a model's request becomes a real effect, drawn in eleven frames: the check, the gate, the run, and the part everyone skips.
- Stage 05 · Human judgment · 10 frames
Human-in-the-Loop Is a System Design Choice
An Approve button isn't oversight. Ten frames on what a person needs before their judgment actually controls anything.
- Stage 06 · Evaluation & evidence · 10 frames
What Counts as Good, and How to Check It
Every check is standing in for a judgment someone wrote down first. Ten frames on naming the property before choosing the method.
- Stage 07 · State & continuity · 8 frames
What Has to Survive for Work to Continue
A transcript records how the work got here. The next person needs where it stands and who owns the next step. Eight frames on the difference.
- Stage 08 · Readiness · 13 frames
When Is an AI System Ready for Real Use?
Tested, approved and deployed are not the answer. Thirteen frames on readiness as a judgment about one exposure, for one use.
Storyboards are drawn one at a time, when an article earns one. Every idea is already written up in Systems.