Field Notes
What the Army Taught Me About Rolling Out AI
Adoption is a leadership problem before it's a technology problem. What twenty-four years of watching the Army re-equip and retrain itself predicts about your AI rollout.
Every conversation I hear about rolling out AI is about the technology. Which model, which vendor, which security review, which license tier. Those are real questions. They are also the easy ones. Almost none of them will decide whether your rollout works.
I spent twenty-four and a half years inside a giant change-management laboratory. The U.S. Army re-equips, re-doctrines, and retrains itself constantly, at a scale no company approaches, with people who did not apply for the change and cannot quit that afternoon. Some of those fieldings took. Many quietly didn’t. Watching the difference up close, from the motor pool to the command group, taught me most of what I know about adoption — and it maps onto AI almost embarrassingly well.
Here is what the pattern predicts.
The middle decides
Every rollout has enthusiasts and it has sponsors. Neither group decides anything.
In the Army, a new system lives or dies with the sergeants — the people who have to run it on a rainy Tuesday when the network is down and the mission hasn’t moved. If the NCOs decide the new thing helps them accomplish the mission, it becomes how the unit operates. If they decide it’s one more thing the staff invented, it becomes a checkbox — technically fielded, functionally dead.
Your company has the same layer: the team leads, the shift supervisors, the senior analysts. They will not attend your AI town hall and they will not argue with your strategy deck. They will simply decide, quietly and collectively, whether the tool becomes practice or shelfware. And unlike soldiers, your people can leave — which makes this layer more decisive, not less. The typical rollout never identifies these people, let alone recruits them.
Field to the middle first. Put the tool in the hands of the people the rest of the organization actually copies, give them real influence over how it’s used, and make them the trainers. An hour of a respected supervisor saying “here’s how I use it” is worth forty hours of vendor enablement.
Train tasks, not tools
Army training doctrine has a discipline that most corporate enablement lacks: you train a task, under conditions, to a standard. Nobody gets trained on “the radio.” They get trained on calling for a medical evacuation, at night, in nine lines, until they can do it stressed and tired.
The standard AI training is the opposite — a tour of the tool. Here’s the interface, here’s a prompt, good luck. Then we’re surprised when usage spikes for a week and dies.
When I built TIMON, a workforce training platform for Disney’s distribution operation, the hard part wasn’t the technology. It was the task inventory: every job, broken into the tasks that make it up, each with a standard. Train the task, the new way, to the standard — and the tool stops being a novelty and becomes how the work is done. Do the same with AI. Don’t teach “prompting.” Teach your intake process and your weekly report — done the new way, to a standard you can name.
The organization watches what leaders do, not what they sponsor
In the Army you learn quickly that soldiers ignore what you say and study what you do. Where you show up, what you check, what you tolerate when you’re tired — that’s the actual policy.
AI rollouts fail this test constantly. The executive sponsors the initiative, funds the licenses, records the kickoff video. Then they run their own decisions and their own week exactly as before. Everyone notices. The signal lands with perfect clarity: this is for you, not for us. Adoption follows the signal, not the memo.
If you lead the organization, use the tools where people can see it. Reference what they produced. Show your own failed experiment in the staff meeting and what you changed. You don’t need to be the power user. You need to be visibly in the fight.
If the pilot can’t fail, the rollout can’t succeed
The Army’s best cultural export is the after-action review: what was supposed to happen, what actually happened, why, and what we change. Rank comes off. The point is truth, because the alternative to truth is casualties.
Most corporate AI pilots are structured so they cannot fail. Success is declared in the same deck that proposed the budget. The teams who found the tool slower, wrong, or risky learn to keep that to themselves — and the rollout scales the fiction.
Run the pilot like an AAR instead. Decide before you start what “worked” means, in numbers you can’t negotiate with afterward. Reward the team that reports “it slowed us down 20% on this task” — they just saved you from scaling a mistake. An organization that can’t hear bad news about a pilot will eventually get very bad news about a deployment.
Installation is not adoption
The thread through all of this: technology arrives by installation, but capability arrives by formation. People, practiced in new tasks, led by leaders they can watch, inside a culture that can tell itself the truth. The Army spends astonishing effort on that second part, because it learned the hard way that fielding equipment without forming the force just gives you expensive storage.
Your AI rollout will be decided the same way. The model is the easy part. The middle of your organization is where the whole thing is won or lost. It was true of every fielding I watched. It will be true of yours.