Beyond the four days — cases for the book · Security, performance and command — the reserve case
Subtitle: What an army knows about performance that most companies have to learn twice
The exercise begins at 04:40, and the first thing that fails is not a machine.
Twenty-two people are in the room: eleven officers on a command course and eleven executives who were invited because the academy wanted to see what happens when the two groups are given the same problem. A port matters to both of them. A storm, a cyber incident at the terminal operator, a ship that cannot be unloaded, a supplier who is already calling, and one piece of information that turns out to be wrong.
They have the same tools. An AI planner reads every feed and produces, in forty seconds, three courses of action with timings, costs and second-order effects. It is good. Nobody in the room could have done that in an hour.
By 05:20 the two groups have diverged, and not in the way the academy expected.
The executives are still comparing the three options. They want the fourth. They ask the planner to re-run it with new assumptions, then again. Their spreadsheet is beautiful and their decision is late.
The officers pick the second option in nine minutes — not because it is the best, but because it is good enough, reversible and someone owns it. Then they spend the remaining time on something the executives never touch: what happens when this decision turns out to be wrong, who will notice first, and what that person is allowed to do at 05:29 without asking.
At 06:10 the injected information changes. The wrong piece is corrected. The officers' choice is now the weaker one. The lieutenant who was told to watch for exactly this says so out loud, in front of the commander, in eleven seconds. The plan changes. Nobody is punished, nobody apologises, and the room does not slow down.
In the executives' group, the same correction arrives at the same moment. It reaches a manager who has spent the night defending the analysis. He needs four minutes to say it, and when he does, the first question he is asked is whose number was wrong.
The exercise ends at 08:00. Both groups miss the ship.
The debrief is where the case actually lives.
The officers' instructor does not ask who was right. He asks three questions in this order: what did we decide, when did we know it was wrong, and how long did it take from knowing to changing. Then he says the sentence the executives write down: performance is not the quality of the first decision; it is the time between the mistake and the correction.
And then the harder half. The academy has been using the AI planner for two years. The officers are faster with it than without it — but only in the tasks they already knew how to do. In the ones they had never trained, the planner makes them more confident and no more right, and confidence is the one thing an unprepared unit cannot afford.
One of the executives asks the obvious question: so should we train our people the way you train yours?
The instructor's answer is the uncomfortable one. Not the drills — you cannot drill a sales team at 04:40. But three habits, and the army has them because it pays for not having them: everyone in the room knows who decides; everyone is allowed to say the thing that ruins the plan; and someone is always assigned to look for the evidence that the plan is already wrong.
The executives have the tools. Most of them do not have the three habits, and no tool will sell them one.
At the end of the debrief, the commander adds a line that changes the room: in his world, the machine is now also a participant — it recommends, it drafts, it watches feeds, and one day it will act. His only question is the same one the officers train: what is this participant allowed to do at 05:29 without asking, and who notices first when it is wrong?
The room will want to make this a story about the military being better. It is not. The officers are not smarter; they are trained for the one hour that decides everything, and they have paid for the habits the executives are still calling culture. Do not let the table settle on "we need more preparation" — ask each person for the number in question 1 and write the numbers on the wall.
The NEO world puts a second kind of participant in that room. An agent can watch the feeds no human can watch, hold the assumptions, and say at 06:10 that the plan is already wrong — the lieutenant's job, without fatigue. It can also do the opposite: produce a fourth option, and a fifth, and keep a room busy until the ship is gone.
Which of the two it becomes is a leadership decision, not a technical one. It depends on what the agent is asked to do (decide, or attest and warn), on what it is allowed to do without asking, and on whether anyone is accountable for reading what it says. The rail attests; it never decides. The same rule, in the room, applies to the agent.
Performance is not the quality of the first decision. It is the time between the mistake and the correction — and that is the one number a machine cannot improve for you.
A question for the table, a disagreement, what you would have done. The case lead reads every comment; the ones the table takes up enter the chapter as questions from the room, with your name.