Engineering
Choices
A Leader's Guide to AI at the Edge
How organizations in the 21st Century can ensure they always have a next move.
The historical argument. The mathematical framework. The AI architecture.
Each one stands on its own. Together they make the complete case.
On the afternoon of August 2, 216 BC, the Roman army marched into a valley near Cannae — and into the most efficient killing machine in military history. They had numbers, discipline, and a plan. What they did not have — when Hannibal's lines curved around them and the pocket closed — was anywhere left to go.
One way of answering the question of what happened is that each soldier's Ω was allowed to fall to zero. Not all at once. One by one. The soldiers on the outer edge still had transitions available. The soldiers deeper in the pocket did not.
The mechanism is identical in modern enterprise. The organization that has optimized away its alternatives — in pursuit of efficiency, one reasonable decision at a time — discovers the same thing when the disruption arrives. Not a failure of management. A failure of optionality.
Nassim Taleb identified the philosophical stakes in Antifragile. The optimized system is the fragile system. This book attempts to add the operational answer to the question Taleb leaves open: not just what kind of organization to be, but precisely how to build one.
"They were the ones who still had choices when the moment arrived."— Engineering Choices, Preface
"Taleb showed us what to want. This book attempts to show you how to build it."— Engineering Choices, Preface
"He didn't predict what would happen at Cannae. He engineered the conditions under which he would always have a next move."
Engineering Choices — The Moment of Choice
Twelve Chapters. Two Thousand Years of Evidence.
Eighteen Human Stories.
Each chapter opens with a historical story — from Cannae to the Wright Brothers — that makes the framework's argument before the mathematics begin. Click any chapter to begin reading.
The Divergent Markov Chains Framework
The book rests on three foundations. Together they provide a complete and operational answer to the question of how organizations maintain strategic freedom of movement.
The objective function Ω provides a precise, computable measure of organizational optionality — how many choices exist, how long they persist, and whether they are eroding before anyone notices.
Edge-deployed Small Language Models run the Ω framework at the operational node in real time — at the speed of disruption rather than the speed of planning cycles.
From Cannae to the 2008 financial crisis, the organizations that survived disruption were never the ones that predicted it. They were the ones who still had choices when the moment arrived.
A continuously computable measure of organizational optionality — combining immediate option quality, horizon persistence, and decay rate — weighted by the organization's current risk posture.
Nelson de Sa e Silva solves problems in enterprise technology environments. The mathematical foundation in this book — the Divergent Markov Chains framework and the Ω objective function — is just his mathematical expression for keeping options open.
When he is not thinking about transition matrices and horizon divergence scores, he is playing guitar or taking computers apart to see what makes them work.
Nelson grew up in rural areas out west. He is what is called in the vernacular a lifetime listener, first-time caller, and has listened and learned from people who were generous with their wisdom. May this book be an artifact of that experience.
Engineering Choices: A Leader's Guide to AI at the Edge is available to read in full on this site.
Registered with the US Copyright Office.
© 2026 Nelson de Sa e Silva.
More in the Print and Kindle Editions
The full text of all twelve chapters is available to read here. The print and Kindle editions include three additional appendices not on this site.
A complete formal treatment of the framework — the divergence score H(i), the horizon divergence score D(i,n), and the Ω objective function, with proofs and worked examples. Includes a reference section situating the framework in relation to Markov Decision Processes, robust MDPs, real options theory, and maximum entropy reinforcement learning. Written as a standalone reference for readers who want to implement the framework precisely.
The historical cases from the main text retold from the inside — with attention to what the people involved knew, what options were available to them, and how those options narrowed. Cannae. The Irish Famine. The Nika Riots. Apollo 13. Lehman Brothers. Bhopal. Several additional cases not in the main text are included. The appendix can be read independently.
A step-by-step guide to building the edge inference node described in Chapter 7. Covers hardware selection, GPU tradeoffs at current price points, software stack, quantization levels, and model feature requirements. Each decision is worked through using the Ω framework. Assumes no prior experience with local AI deployment.
Nelson is around sometimes. You should see if you can get in touch with him.
Find him on LinkedIn.