A Book Every Four Days
At some point this spring I got asked, more or less, whether my vocabulary was mine.
Not rudely. It was a fair question. I had been writing about judgment, failure modes, blast radius, load-bearing decisions, the difference between a system that can be argued with and a system that cannot. I had been writing it with an assistant in the loop. And when a tool that fluent is in the loop, the question of who thought of what stops being rhetorical. I had already caught it once handing one of my own ideas back to me as its own, and I could show where it got it. So I went looking for the other direction. Where did I pick it up?
The answer turned out to be in my Audible history, and it was more specific than I expected.
The Cadence
From the middle of October 2025 to the middle of May 2026 I listened to fifty-one books. That is one every four days for seven months, around a full-time architecture job and a house with kids in it.
I did not notice the cadence while it was happening. I noticed it when I laid the list out with dates next to it and saw that it was not a reading habit. It was a curriculum, and nobody had assigned it.
The first two months alone: Gene Kranz on Mission Control. Tracy Kidder on a computer team under deadline. Langewiesche on fly-by-wire. Junger on a boat that did not come back. The crash investigators. The Halifax explosion. Deepwater Horizon. Chilean miners. Three wildland fire books in a week. The space shuttle. Matthew Syed on why aviation learned from its failures and medicine did not.
Twenty-two books between October 15 and December 17, and almost every one of them is the same story told in a different uniform. A system meets a force it cannot talk its way past. Somebody competent answers the question they were asked instead of the question that mattered. The record afterward shows exactly where it went wrong. The record is the only thing that improves the next one.
Failure autopsies with a plot. That is what I read for. I did not have those words until I saw the list, and I am a little embarrassed by how consistent I am.
The Eight Days
The part that made me sit up was a stretch in late January.
There is a podcast called Causality that takes apart engineering disasters one episode at a time. Sometime before January 27 I listened to the Therac-25 episode. A radiation therapy machine whose software killed patients because of a race condition. The organization around it could not see the race condition, or would not admit that it saw it. Every theme I had been reading toward for three months, compressed into one technical case with a body count.
I brought it up in a conversation the same day and called it the moment something clicked. That is my word, from the transcript, not something I went back and added later.
Two days later I finished the 737 MAX episode. Eight days after the click, on February 6, I listened to Meltdown, which is the readable version of Perrow's normal-accident theory. The same day, in a completely different context, I wrote that I formally document the risks we are accepting and have zero problem taking responsibility for the decisions I made. That sentence is the seed of everything I have published since. I did not know it was the same day until I lined up the timestamps.
I am not claiming cause. I am claiming that the vocabulary I was later asked about has a paper trail that starts months before the tools, and the trail has dates on it.
Why It Matters
The lazy version of this worry says an assistant will make you sound smarter than you are. I do not think that is quite the risk. The risk is that it will make you sound like it, and you will stop being able to tell.
The cheapest defense I have found is not a rule about how to prompt. It is a reading practice with timestamps, one that was loading the model between my ears for years before any chat window existed. When I read my own sentences now and wonder whether the thought was mine, I can go check what I was listening to that week. Usually the book got there first. That is not proof the thought was original. It is proof of where it came from, which is the thing that was actually in question.
The books do not make the judgment. They give it shape and words, and they let me see a problem from the bottom up before I touch it. I have a bottom-up brain. The bookshelf is how it got stocked.
The List, Roughly
People ask, so here it is. Perrow's Normal Accidents and Adrian Newey's How to Build a Car, both before the run started. Then the fall: Failure Is Not an Option, The Soul of a New Machine, Fly by Wire, The Perfect Storm, The Crash Detectives, Trapped Under the Sea, Fire on the Horizon, Deep Down Dark, Black Box Thinking, Young Men and Fire, Fire on the Mountain, Into the Black. Winter: Five Days at Memorial, The Control of Nature, A Crack in the Edge of the World, Meltdown. Spring: The Box, Underland, Fire Weather, The White Cascade, Eruption, both books called Chaos, These Strange New Minds, and, started the day after I stopped counting, Designing Data-Intensive Applications, which is the one people expected to be on the list all along.
Fire on the Mountain is the best of them. If you only read one, read that one, and then notice what it has in common with a codebase you are afraid of.
-- Justin Higgins. Software Engineer, Midwest. Fifty-one books about things that broke, then a summer building things that refuse to.
Companion pieces: The Work Looks Like Nothing - chat history is downstream of the bookshelf. The Tree I Never Planned to Move - the other direction of the provenance question.
Reactions, disagreements, war stories: jchigg2000.dev@gmail.com