Preface: The Job Is Now the Person
Count what you signed this week.
The analysis you approved after reading the summary instead of the data. The spec that went to engineering with sections you did not write and did not entirely read. The competitive brief built from filings you never opened. The recommendation you accepted because it was fluent, well structured, and arrived in four seconds, and because the last forty like it were right.
None of that makes you careless. It makes you current. The tools are real, the speed is real, the quality is mostly real, and you are producing more than you have produced at any point in your career. You probably feel sharper than ever. Most people working this way do.
Now one question, and the whole book is in it. When the machine is wrong, quietly wrong, fluently wrong, wrong in the exact format and confidence of all the times it was right, are you still the person who can tell? Not in principle. This week. On the forty-first recommendation.
And the harder question underneath it: how would you know? Because the feeling of being sharp is not evidence. The research this book opens with says the feeling and the fact have come apart: confidence rises with every assisted week while the independent capability quietly falls, the decline is invisible to the person declining, and the people most certain they are exempt are, measurably, the most exposed. The skill that is supposed to catch the machine’s mistakes erodes through exactly the use that makes you productive, and it sends no signal when it goes.
Look around at what else you depend on. The systems you ship have dashboards and alerts. Your data has lineage. Your code has tests. Your car has service intervals; your teeth get checked twice a year. Every asset anyone depends on has a maintenance program, except the one that signs off on all of it. Your professional judgment, the thing every decision in your week routes through, is maintained by nothing but the feeling that it is fine, and the feeling is broken in the direction of reassurance.
This book is the maintenance manual.
It is a practice, not an argument, and it runs in four parts. The first lays out the stakes: the evidence that judgment is perishable, that oversight ability is a skill rather than a credential, and that the apprenticeship which used to maintain all of this for free is being dismantled, feature by helpful feature. The second rebuilds that apprenticeship deliberately, for one practitioner at a time, ending with a proficiency regime that proves your judgment still works instead of trusting that it does. The third puts the maintained judgment to work at the three decisions nobody can take from you: what to build, whether to ship, and how to be heard while saying so. The fourth asks what all of it is worth, in a job market that has not yet learned to price or certify the one capability it most needs.
One thing the book does quietly, and you may as well know going in. Nearly every chapter leaves something behind: a template, a checklist, a log, a file. None takes long to build. By the last chapter they turn out to have been the point, because in a profession with no license and no recertification, the only credential that means anything is the documented state of your own practice.
A word on evidence, because the book will hold itself to it. Where serious research exists, I use it, and the studies in the first chapters deserve your full attention. Where it does not exist yet, and for most of the practice it does not, I say so plainly and hand you the tool anyway, as a designed practice you run, measure on yourself, and adapt. Product management is not evidence-based medicine, and waiting for the randomized trial of judgment maintenance means arriving at the future unmaintained.
This is the fourth book in a series about agentic AI products, and it stands alone; the single page after this preface carries every term it borrows, and you need nothing else from the earlier volumes. They built the operating loop, the failure catalog, and the team. To readers who finished the third book: nothing here walks back its argument. The team rightly took the work. It cannot take the judgment, and no seat on any org chart maintains yours.
I came to product management from medicine, a profession that has been working next to a confident machine for longer than most product managers have been alive, and the practice in this book begins with a game every medical student learns without ever being taught.
The job did not shrink when the machines took their share. The job is now the person.
This book is the maintenance manual.