This Echo is a little different from the previous ones. It does not tell a story or share a reflection. It announces a series of essays that may eventually become a book. The Almost-Right Machine: Adlerian Essays on Software Engineering in the Age of AI will be published here, one essay at a time, and this page will be its home. Several of the fifteen essays are already written, but are still under review. I will release them one at a time, and write the rest along the way.
Fifteen essays, written in the open
The form is borrowed from Fred Brooks, whose The Mythical Man-Month is a set of essays rather than one long argument. Each essay starts with a scene, real or imagined, in a classroom, a conference, or a code review, and only then reaches for the idea. Each stands on its own. Each will get a short Whisper here as it is published: the scene it opens with, the sentence it turns on, and one question for you. The full text lives in the book.
- The Almost-Right Machine
- The Skill and the Discipline
- Adler in the Machine Room
- Three Ways to Lose
- The Fiction of the Fast Hands
- Whose Task Is This?
- The Honesty Defect
- “The AI Generated It”
- Belonging Without Ranking
- The Silent Teacher
- The Conditions of Striving
- Before the Defenses Form
- The Oral Defense
- No Silver Engineer
- Who Is Being Formed?
The essays themselves are on Leanpub, free, with each new one added as it is written.
You can read them in any order. If Adler is new to you, start with the third essay, the only one that explains him. If you manage engineers, start with the eleventh. If you teach, the twelfth and thirteenth are yours. If you build the tools, the tenth. If you have met the almost-right machine and did not know what to call it, start at the beginning.
Almost right
The title comes from a student. In my graduate course on trustworthy software engineering, his team spent a week deliberately injecting faults into the system they had built, with a machine writing most of the code, to see which of their own checks would notice. While looking for something else, they found code whose names, comments, and metadata described three features, and which computed none of them. It was complete, fluent, internally consistent, and empty.
At the end of the semester, when asked what the course had taught him, he wrote that AI is most dangerous when it is almost right. Every engineer has said some version of that sentence. He put it where it belongs. Crashes and syntax errors are obvious. The hardest defects are the ones that look finished.
Both sides of almost
For twenty-five years I have asked one question: not just whether a system works, but whether it deserves to be trusted. My field measures the gap between what a thing claims and what it does. The almost-right machine does most of what you need well enough that you stop checking, and that gap has quietly become the central problem of the profession.
But the phrase cuts the other way too, and that is why it names the essays. Our responses to the machine are also almost right. Dismissing it is almost right, because its flaws are real. Surrendering to it is almost right, because it is genuinely capable. Handing it your judgment is almost right, because it is fast and you are tired.
Each looks like a finished answer, and none of them is. These essays are about the difference between an answer that looks finished and one that is, in the machine and in the person using it.
Why Adler
The lens I use is an unlikely one. I came to Alfred Adler by accident, in a bookshop in Rio de Janeiro, and kept reading because he helped me name something my field has no vocabulary for: how an engineer feels when the machine is better than he is, and what he should do next.
Adler’s claim is simple to state and hard to accept. We all begin helpless, and much of what we do afterward is a response to feeling less than we need to be. The question is never whether we feel inadequate next to something more capable. It is what we do with the feeling: grow, or defend, avoid, and quietly give up.
This is about software engineering, and Adler is the lens, not the subject. It is not a psychology text, not a manual of prompts and workflows, and not a reassurance. When I first tried these ideas on a room of dependability researchers, I expected polite tolerance and got real questions instead. That gave me the courage to write it.
An invitation
I am still early in this, and I expect some of what follows will look different to me a year from now. That is part of why I am writing it in the open. Adler’s answer to the question of worth was contribution: not how good am I, but who is this for. Writing these essays here, where you can push back, is my small attempt to take that answer seriously.
If you have met the almost-right machine, in your code, your team, or your classroom, I would like to hear the story. If you think I am wrong, even better.
Subscribe to get a Whisper for each essay as it is published, or follow the book in progress on Leanpub, free, with every new essay included.