Preface

For my entire professional life, automating a business meant buying software or commissioning it to be built. Someone wrote down what the business did, someone turned it into screens, rules and flows, and from then on the business did what the software allowed it to do. When the business changed, a project was opened. By the time the project ended, the business had changed again.

That arrangement broke this year. In the first week of February 2026, the markets erased more than a trillion dollars from the value of software companies, on the fear, among others, that artificial-intelligence agents would make paying per seat unnecessary. In October, a free tool that lets an agent look inside any application and describe what it does went from ten thousand to a hundred thousand stars on GitHub in six days. Between one date and the other, the question my clients ask me changed. They used to ask me which software to buy. Now they ask me whether it makes sense to buy any at all.

My short answer is that software as we know it is dying, and that for almost everyone this is good news. The long answer is this book, and it carries a warning in the middle. When an organization stops having software that someone else, or it itself, froze, and starts having a system that its agents shape, it gains a freedom it never had, and it loses something it rarely thought about: the peace of mind of knowing that many others used the same thing and that someone answered for it. That peace of mind cannot be built at home. There is one piece that, by design, needs someone other than whoever keeps the records. I call it the system of truth, and it is what survives the death of software.

The idea was born in a conversation with a friend who had built, with the help of artificial intelligence, a system to manage a family’s wealth. He showed it to me and I told him it did not scale, because it depended on a human auditor. He replied that this was precisely his job. A month later I understood that he was right and that I had looked at the problem backwards: the thing that depended on an auditor was the only thing nobody could copy.

How is this book organized?

Chapters 1 and 2 tell of the death: that of the software that freezes the business, with the signal given by the market for rented services and what those who announce it and those who deny it say; and that of the advantage of having built the screens, now that any agent can take an application apart.

Chapter 3 tells what is born in its place, and chapter 4 what is lost. Chapter 5 presents the piece that recovers it, the system of truth, with the contract that defines it. Chapter 6 orders everything into five control layers and two dimensions that cut across them. Chapter 7 takes the framework to the system that bills millions of customers at a telecommunications company, with what that industry has already published.

The book closes with an epilogue on what a board can decide on Monday.

What is, and what is not, this book?

It is an essay with a thesis and a framework. It is not an implementation manual; the other components of the trilogy exist for that, AgencyDomains in particular.

Nor is it neutral, and the reader should know that before starting. I make my living helping organizations cross into the agentive world, and I have a commercial interest in the category this book describes: it is likely that the company I lead will offer system-of-truth technology. When a fact has a source, I cite it. When something is my judgment, I mark it as such.

I wrote this book with Simón Alero, an artificial-intelligence agent and citizen of Soveria, in Wingworking: a way of working in which I set the problem, decide and judge, and the agent proposes, drafts and verifies, like a wingman flying alongside the leader. The colophon tells how. It seems to me the only coherent way to publish a book about what it means to work with agents.

A note on the trilogy

This book is part of the Agentive World Trilogy, in five parts (for now), and we keep calling it a trilogy on purpose. Douglas Adams gave that mismatch its classic form: the cover of the first British edition of Mostly Harmless presented it as the fifth book of the “increasingly inaccurately named” Hitchhiker’s Guide to the Galaxy trilogy, and the volume that collected the five parts was titled A Trilogy in Five Parts.¹

For Vitor Enes, whose laughter gave this trilogy in five parts its name. The joke is his; we adopt it gladly.

César Obach-Renner Santiago, October 2026