Architecture

Systems

Eight questions an applied AI system has to answer, in the order the answers depend on each other.

The systems map

Where are you in the system?

Each stage opens the explanation that deals with it. Follow the sequence, or enter where the question becomes useful.

01How models work
What is actually generating the output?
What an AI Model Actually Is Storyboard
Goes deeperFrom Text to Tokens Training, Fine-Tuning, and Inference
02Context & retrieval
What can it see, and how did that get there?
Retrieval-Augmented Generation in Plain Terms Storyboard
Goes deeperContext Windows: What the Model Can See in One Run Why OCR Quietly Breaks Document AI
NotesSemantic Caching: Similar Is Not Safe to Reuse
03Instructions & contracts
What are we asking it to do, and what is enforced?
Prompting Is Not the Skill You Think It Is Storyboard
Goes deeperStructured Output and Why It Matters What a System Prompt Actually Is Prompt, Skill, Workflow, Agent: Telling the Layers Apart Designing Reusable AI Skills
04Action & authorization
What can become a real effect?
From Agent Intent to Governed Execution Storyboard
Goes deeperTool Use: When Language Triggers Actions AI Agents vs AI Workflows What Is an Agent Harness?What Is MCP? The Model Context Protocol in Plain Terms Policy-Governed MCP Runtimes: Trust Boundaries for Tool Execution
NotesWhy the Loop Outlives the Model
05Human judgment
Where does a person have to decide?
Human-in-the-Loop Is a System Design Choice Storyboard
Goes deeperWhen Chat Renders an Interface What Is Bounded Autonomy?
06Evaluation & evidence
How do we know the behaviour was acceptable?
What Counts as Good, and How to Check It Storyboard
Goes deeperWhat to Record So You Can Explain What Happened
07State & continuity
What has to survive across a boundary?
What Has to Survive for Work to Continue Storyboard
NotesThe I-7 Loop: A Working Method for Delegating to AI The Instruction Spec: Defining an AI-Assisted Software Change
08Readiness
Is there enough evidence and control for this use?
When Is an AI System Ready for Real Use? Storyboard
Goes deeperWhat Does Forward Deployed Mean?What Is a Forward Deployed Consultant (FDC)?

Across the map

Not one stage, but all of them.

Overview

How the Parts of an AI Application Depend on Each Other

The boxes in an architecture diagram are the easy part. What matters is what each part establishes, and what the next part is assuming was already settled.

Note

What Systems Thinking Helps You Inspect

A handful of old ideas that still earn their place when looking at an AI system: where the boundaries are, what the feedback does, and three ways things go wrong quietly.

Note

What a Publisher Actually Controls

What SEO, AEO and GEO mean: search, answer and generative engines. They describe someone else's system; a publisher controls a short, checkable list.