Epilogue · What Dies, What Is Born and What Survives

What does this book tell?

It tells that one way of automating a business is over, and that what comes in its place needs a piece almost nobody asks for yet. I repeat it here in one go, because the chapters said it in parts and it is worth seeing it whole once before deciding anything.

For half a century, automating meant describing the business once and freezing it in a piece of software, whether it was rented as a service, licensed and operated in-house or commissioned to be built. The arrangement made sense as long as writing code was expensive and changing it slow. Both stopped being true, and in 2026 the market noticed twice: in February, when it erased more than a trillion dollars from the value of software companies, and in October, when a free tool put within reach of any agent the ability to look inside an application and describe what it does. What dies is not all software. What dies is the frozen business: the one that changes only when someone opens a project.

In its place, and this is my thesis, organizations are born that have a system their agents shape and change as the work arrives. It is not that an agent programs the same application faster; it is that nobody programs it in advance. The agent attends to the work by reasoning, crystallizes what repeats into deterministic pieces of its own that resolve it the same way every time, and confines into bounded, measured pieces what asks for judgment in every case. Determinism, which is rightly demanded of any system, does not disappear: it survives in the core that is still bought, and in the work the agent attends to it stops being written by someone in advance and becomes what the agent produces when it learns.

Those organizations gain a freedom they never had and lose something they did not write into the budget: the peace of mind that many others used the same thing and that someone answered for it. They lose it without noticing, until someone, at a board table, asks whether the numbers that come out of the system are true. Building well does not answer that question, because it is not a question about quality but about who vouches, and nobody vouches for themselves.

It is answered by a piece nobody should certify for themselves: the system of truth, a system of record that, besides keeping the official version of each piece of data, guarantees that what is recorded is true through a contract of six properties and two guarantors, the technology vendor and the one who attests, independent of whoever records. It is the piece that survives the death of software, and it is what is still bought when the rest is no longer bought. Around it go four more control layers, identity and authority, movement, measurement and judgment, cut across by two questions asked of each: who operates it and how to go back. The telecommunications operator of chapter 7 shows that all of this already has a name in the most demanding industry, and also that this industry trusts its artificial intelligence quite a bit more than it can demonstrate.

That is what the book tells. What follows is what can be done with it.

What can a board decide on Monday?

A board does not need to settle the future of software to act on Monday. Three decisions are within its reach without waiting for anyone.

It can ask about the truth before asking about the product. In the next decision to buy, renew or replace a system, the first question is which records it holds that the organization is accountable to others for, and who vouches for their being true. “The vendor, because many use it” is a good answer, as long as it is known that it is being paid for. “We do” is an incomplete answer.

It can ask every offer for the five control layers, the suite, the custom development and the construction of one’s own: how it answers for the truth, who acts and on whose behalf, how far its agents decide, how it is measured that they do it well, on which standards it is built, and for each layer who operates it and how to go back. An offer that leaves any one of those questions unanswered cannot yet be compared.

And it can adopt a short rule: what cannot be shown to a third party does not count as a control. The telecommunications operators of chapter 7 trust their artificial intelligence quite a bit more than they can demonstrate, and a board has no reason to inherit that gap.

The rest of the trilogy tells where this transition is going, which way it is traveled and how it is built. This book wanted to answer why one has to cross now, and what cannot be left behind.