Chapter 5 · The floor and the plateau
The question Marcus asked on Friday
Marcus Ellery asked one question at the end of Priya's demonstration and it emptied the room's good mood in about four seconds.
He wanted to know how long it would take the distributor in Toledo to do the same thing.
Nobody had an answer, because nobody had thought about it, because the demonstration had been so obviously impressive that its being easy had registered as a virtue rather than as a fact about the world. Priya, who is honest to a fault about her own work, said probably an afternoon, and that a competent person who had never done it before would need a day.
Marcus said that in that case he was not sure what they had just watched.
He was right, and the thing he had put his finger on takes a while to see clearly, because it runs against how improvement normally feels. Two years earlier what Priya did on that Thursday would have been a quarter of engineering time and a vendor contract with a statement of work. That collapse is real and it is enormous. It is also the part of this story that will matter least to Ostermill by the last page.
The reason is that it happened to everyone on the same Thursday.
The distributor in Toledo gets the same afternoon. So does the competitor twice Ostermill's size with a receivables team of thirty, and the one half its size that has been undercutting them on hydraulics since 2023. Capability that arrives for everybody is not an advantage. It is a floor, and the floor rises whether or not you personally step onto it.
Ostermill did not win anything in March. It stopped being behind.
That distinction disappears constantly and it disappears in one direction. A team measures the improvement against its own past, sees a real gain, and books it as a competitive move. The comparison that decides anything is against the same competitors in the same quarter, and on that comparison the gain is usually zero, because the capability was bought rather than built and the seller has other customers with the same problem.
Where the climb stops
There is a second effect underneath the first, and Dana had watched it twice before she arrived at Ostermill.
Tools that make everyone competent in an afternoon also make almost everyone stop at competent. The first afternoon takes you from nothing to a working demonstration, which feels like the bottom of a slope you could keep climbing. The second afternoon takes you from a working demonstration to a slightly better working demonstration. The fifth produces something you would not show anyone. The tenth produces nothing at all, and by then the team has quietly concluded that this is simply what the technology does.
At her previous employer the conclusion had arrived in about six weeks, and it had been wrong both times.
The climb flattens because the tool was doing the climbing, and the tool has the same ceiling for everybody who buys it. Nothing above that ceiling is prompt-shaped. Nothing above it appears in a release note.
What sits above it is the work Priya did not do. In the last chapter's terms it is the boundary, the supervisor, and the two halves of the context she only gestured at. Deciding what the boundary is and whether it is a sentence or a wall. Working out which source wins when the aging report and the remittance advice disagree, which they do, weekly. Sitting with Ruth long enough to get her reasons out of a woman who has never once been asked to explain them. Building the thing Marcus will look at on a Monday to find out whether any of this is working.
Every one of those is unglamorous, none of them demonstrates well in a Friday meeting, and all four are the actual product.
This is also the answer to something the last chapter left open. Dana's second failure shape, the crawl-walk-run roadmap whose "walk phase" is never staffed, gets explained by escalations and migrations and people leaving, and all of that is true. It is not sufficient. The walk phase loses every argument it is in, because the thing it would fix has not visibly broken yet, and hers lost.
The plateau is why. By the time the walk phase comes up for staffing, the crawl version works, the last three afternoons of effort produced almost nothing anyone could point at, and everybody in the room has quietly updated toward the view that this is roughly as good as the technology gets. There is no case left to make. A roadmap of that kind does not get cut in a meeting. It gets outlived by a belief nobody remembers forming, and the belief is an artifact of the tool's ceiling rather than a fact about the work.
Ostermill's competitors are on the plateau right now. So is Ostermill, in the first week of March, and the only reason it does not stay there is that Marcus asked an awkward question and Dana had seen the answer twice.
What is worth being good at
There is a personal version of this, and it is the one people skip, because the floor is where the learning feels fastest.
Skill with a particular tool depreciates fast. The thing Priya used that February will be superseded, its interface will change twice, and the specific technique she worked out for holding a tone policy in context will be irrelevant inside two years. Time spent becoming excellent at it earns a return for roughly as long as the release cycle allows, which is not long enough to build a career on and is exactly long enough to feel like progress while you do it.
Skill at describing judgment does not depreciate.
Knowing how to sit beside Ruth for a week and come away with something written down is a capability that survives every model generation, because what it operates on is the company rather than the technology. So is knowing which boundary needs a wall. So is being able to read a fluent, confident, correctly formatted letter and see that it should never have been written, which requires knowing that the owner died in July and cannot be bought from anybody.
What sits above the floor is the same work it always was, and fewer people are doing it now than were doing it two years ago, because the floor is comfortable and the view from up there looks a great deal like progress.
Which leaves those four jobs, and they are not equally hard. The boundary, the precedence rule, Ruth's reasons for the forty-one, and Marcus's Monday view. Two of the four are context, and context decides how much the other two have to carry, which is why it is the expensive one and why it was the one Priya gestured at and went no further into. The demonstration worked without it. Nobody in that room had asked the model a single question about Ostermill.