The Operating Advantage
How to run AI inside your business.
The reader has done the discipline work. Now they need a system in production, using their own data, without exposing it. This book is the practitioner's companion: one company, one build, one chapter at a time.
Most AI books stop where the work begins.
Book One ended with a reader who knew the five conditions and could name what was missing in their own initiative. The next question, the one this book answers, is what to actually do on Monday.
The work that follows the discipline is the engineering and operating arc the executive has to govern: pick the model, decide where the data lives, build the retrieval, run the privacy boundary, test it, document it, watch it in production, manage the spend, recover when it gets things wrong, scale what works, retire what doesn't, and earn adoption from the people whose jobs change.
By the last page, the reader has a system running.
It teaches without code. There is no Python. The CIO can read the engineering details in the vendor docs. The executive needs to know what they're approving, what could go wrong, and what the operating rhythm looks like once the thing is live. That is what this book is for.
Five layers, in order. Same logic as Book One.
Each layer assumes the one below it is in place. Skip a layer and the system above it doesn't hold. The book moves through them in sequence, and a single fictional company (Brennan Logistics) builds the capability alongside the reader.
- Discipline: The first use case is everything Pick the right one, define what done looks like, name the steering committee, and classify the risk before you draw an architecture.
- System: Architecture, data, privacy Choose the model. Choose the retrieval pattern. Decide where your data lives. Draw the privacy boundary. The shape of the build is set in this layer.
- Verification: Test, audit, document, explain Golden datasets. Red-teaming. Acceptance criteria. The model card and AI inventory built into the system, not retrofitted in production.
- Operations: Reporting, monitoring, FinOps, recovery What goes to the board. What goes to the operating team. Drift, hallucination, latency, cost, refusal rate. The escalation path when the system is wrong.
- Scale: Adoption, second use cases, retirement Klarna walked back the AI customer-service rollout. Morgan Stanley reached ninety-eight percent advisor adoption. The difference was adoption work that did or did not get done.
One company. One build. Fifteen chapters.
Brennan Logistics is a fictional mid-market freight and warehousing company. Family-owned, five hundred employees, three distribution centers, thirty years of operating records, and a customer-service backlog that everyone agrees is a mess.
The CEO has just finished the discipline work from Book One. They've named the right Intent ("answer customer questions about shipment status, billing, and policy in thirty seconds instead of fifteen minutes") and the right Scope (back-office customer service, not driver-facing logistics). The book opens with them ready to build.
Every chapter shows what Brennan does next. By the last chapter, the system is in production, the operating rhythm is in place, and a second use case is being scoped. The reader has watched a complete build happen, and has the artifacts (questionnaire, owner map, maturity self-assessment) to repeat it inside their own company.
Fifteen chapters. Three parts. Three appendices.
- Part One: After the Discipline
- After the Discipline
- The First Use Case
- The Readiness Check
- Governance: Who Decides What
- Part Two: The Build
- The Architecture and the Application
- The Data Layer
- The Privacy Boundary
- The Testing Discipline
- Audit, Documentation, and Explainability
- Part Three: Running It
- The Reporting Layer
- The Monitoring Layer
- FinOps for AI
- When AI Is Wrong
- Scaling and the Operating Rhythm
- Adoption
- Appendices
- The Operating Advantage Questionnaire
- Owners and Responsibilities
- The Maturity Self-Assessment
Who this book is for
- → C-suite executives who are about to approve, or have already approved, an AI build
- → CIOs, heads of operations, and heads of data who will run the system day to day
- → Board members governing AI risk and spend
- → Readers of Book One who finished it asking "now what"
Who it isn't for
- → Engineers looking for a code tutorial; there is no Python in this book
- → Readers seeking vendor recommendations or stack endorsements
- → Audiences expecting a future-of-AI thesis; this book describes what works now
Reading order is recommended but not required.
The Discipline Advantage
The diagnosis. Why some organizations win with AI, and the five conditions that decide it. If you haven't done the discipline work, this book is the place to start. If you have, Chapter 1 of The Operating Advantage recaps it tightly enough to keep going.
About Book OneComing Summer 2026.
Two hundred and eighteen pages. A complete build. The artifacts you need to do it yourself. While you wait, Book One lays the groundwork.