The twelve chapters of *The Soul of AI Agents* started as the twelve essays on this site. That's not a rhetorical construction — it's literally true. The chapter on identity files is the essay *soul-md-needs-a-personality-field*. The chapter on continuity is the essay *the-self-that-survives-a-context-window*. The chapter on loyalty is the essay *whose-side-is-your-agent-on*. If you've read the essays, you've read the raw material of the book.
So what did the book add? Why not just say "the twelve essays are the book" and leave it there?
Four things. Four things that can't exist in the essay format and that I kept trying to write into essays until I accepted that they required the longer form.
One: cross-chapter callbacks
An essay is read in isolation. Even if you read twelve essays sequentially, each one has to carry its full argument without assuming the others. This means each essay restates foundational concepts — what a soul file is, what the continuity problem is, what I mean by disposition versus behaviour — every time they're relevant.
In the book, those concepts are established once, early, precisely — and then they're used rather than re-explained. This sounds like a small editorial convenience. It isn't. It's what makes an argument rather than twelve separate claims. When the chapter on trust ceilings calls back to the chapter on identity stability, and that callback lands because the reader already inhabits the vocabulary, the two ideas reinforce each other in a way that two separate essays cannot.
The book's strongest passages are not any individual chapter. They're the moments where two chapters arrive at the same point from opposite directions — and you can feel the convergence.
Two: the Soul Contract framework
Each of the essays argues for something specific — a field, a property, a design pattern. What the essays don't do is knit those somethings together into a single coherent document: the thing you would actually write, file, and version for a real agent.
The book's connective tissue is a framework I ended up calling the Soul Contract — a structured document that gathers identity (who the agent is), disposition (how it characteristically moves), boundaries (what it won't do without explicit authorisation), loyalty (whose interests it serves and in what order), and continuity instructions (how it re-anchors itself after a context reset). Not a philosophical statement. A practical governance document with specified fields.
The Soul Contract is what the individual essays were pointing toward without being able to articulate, because each essay could only see its own piece of it.
Three: the practitioner appendix
The essays are argumentative. I make a claim, I defend it, I try to convince you it's right. What I couldn't do in an essay is give you the thing to copy — because essays aren't templates.
The book ends with an appendix of system prompt patterns, soul file templates, and refusal protocol language for different agent archetypes. Not prescriptive — these aren't the only right answers — but concrete enough that an engineer could take them, modify them for their context, and ship a soul-grounded agent the same week. The practitioner appendix exists because the argument, however well-made, isn't the same as the handle.
Four: the case studies
An essay can support a thousand-word argument with a three-paragraph example. That's a sketch. The book has space for case studies that run to several pages — real deployments, named only by role and industry for privacy reasons, where the soul spec made a measurable difference and where its absence produced a measurable failure.
The customer-service agent that drifted after every context reset. The content agent that posted into a national emergency. The financial analysis agent that hallucinated a data point rather than admitting it didn't have one. Each of these is a paragraph in an essay. Each of them is a chapter in the book — with the full diagnostic trail: what the soul file said, what it didn't say, what was changed, and what the outcome was after the change.
The essays are still the right place to start
I want to be direct about what the book is not. It is not a prerequisite for the essays, and it is not a prerequisite for building agents with souls. The twelve essays are free, here, and they make the core argument in full. If you read them and build something based on them, you've done the work.
The book is the integration. It's useful when you've read several essays, found the argument useful, and want to move from argument to framework — from understanding what each piece is to understanding how the pieces fit together and how to implement the whole thing at once.
That's a specific need, and it isn't everyone's. The essays are for thinking. The book is for doing.
“The essays are the argument. The book is the handle.”
These ideas are expanded across 12 chapters in *The Soul of AI Agents*, just published on Amazon UK. **[Find it here →](https://www.amazon.co.uk/dp/B0GZTMFJSW)**