AI Organization & Separation of Powers — Structure Log

An implementation log of organizing AI agents

How to Navigate This Series (A Map of All the Chapters)

2026-06-02

At the end of the previous chapter, I wrote: "From the next chapter on, we will get into the specifics — concrete records of what was actually designed and tested."

That is still true. The next chapter is where the real content begins. But before that, I want to pause just once to show you a map of this series as a whole. Knowing "where this is heading, and where we are right now" tends to change how reading it feels.

Whether you look at the map first is entirely a matter of personal preference. But having the full picture can make it easier to decide things like "I'll skip that chapter and come back to it later." That is why I am laying it out here.

What Chapters 1 through 9 Covered

From Chapter 1 through Chapter 9, this series introduced what it is, what approach it takes, who it is written for, and what it promises.

To be a bit more specific, here is what each chapter covered:

Chapters 1 through 9 together form what I am calling "Part 1: Introduction and Foundations." The intent was to lay out the terms of this series upfront, so that when the main content begins, it does not feel like a sudden shift in direction.

Looking back at it now, keeping each chapter to a single topic seems to have given each one a reasonably clear role.

The Shape of What Comes Next

The main content falls into three broad phases.

The first phase is "Building the Foundation."

This part works through the core terms one by one: AI agent (a program that carries out tasks on its own), separation of powers (in this project's usage: dividing execution, audit, and approval between separate agents — not the legislative/executive/judicial division from civics class), the two-person rule (the principle that irreversible actions require two parties to confirm), the document-first principle (the practice of recording all decisions in writing), and SSOT (short for "single source of truth" — one authoritative reference that everyone uses).

I want to be clear about how "separation of powers" is used here. This is not the familiar concept from school — dividing a government into legislative, executive, and judicial branches. In this project, it refers specifically to separating the roles of "who executes a task," "who checks the result," and "who makes the final call" when AI is doing work. Because this is a specific usage, I will include a note on first use — something like "this is how the term is used in this project" — each time it comes up.

This phase is designed to function as a dictionary for the design records that follow. It is the place to come back to when you find yourself thinking "what did that term mean again?"

The second phase is "A Record of What Was Actually Built."

Here the nature of the writing shifts. The focus moves to process: "designed this," "ran it," "it did not work," "changed it." This is the core of the series.

The topics include the golden rules (the judgment guidelines I developed for deciding how the system should behave), a setup where multiple AI agents are each assigned distinct roles, a mechanism combining a Kill Switch (a setup where the process stops automatically unless a human checks by a set deadline) with automatic publishing, and records of failure. A large part of this phase covers things that did not go as expected.

The third phase is "Scaling the System, and the Question of Accumulated Data."

This covers the thinking behind taking a structure that worked in one project and extending it to others. What it means to keep building up a record over time. And one slightly unusual chapter I am planning: a chapter about the system that is running this very series right now. This series runs on top of the same automation it describes — which makes that chapter a kind of self-reference.

How to Read This Series

You do not need to read in order from the beginning.

If the foundation chapters are what interest you, it is perfectly fine to skip Part 1 and start there. If you are already comfortable with AI terminology, you might find it more interesting to skim the foundation phase and start from the design records.

If you hit a term mid-series and find yourself thinking "I do not remember what that means," the foundation phase has all the explanations in one place. Think of it as a dictionary you can open any time.

Each chapter is written to be self-contained around a single topic, so it is easy to read selectively and pick up only the chapters that interest you. This is a long series, but it is not designed to be read in one sitting — it is meant to be an open record you can come back to when something catches your attention.

To make sure you do not get lost wherever you enter, every time a term appears I will break it down on the spot. And whenever a chapter assumes background from an earlier one, I will add a note along the lines of "this was covered in the foundations phase, if you want to go back and check."

That is enough preamble.

From the next chapter on, the real content begins. What it actually means to design an organization with AI. The record of what I did with my hands — written in order, as it happened.

← cd ..