A note to close out the week.
I spent this week talking with training directors and command staff about AI, and the same pattern kept showing up: everyone is being asked “what are we doing about AI,” and almost nobody is being asked the better question — “what should we be doing about it.”
Those are different conversations. The first one produces a purchase. The second one produces a strategy.
Here’s what thirty years around high-stakes decision-making has taught me, distilled as plainly as I can put it: skill under pressure is not discovered in the moment. It’s built beforehand, in volume, somewhere safe to fail. Every profession that operates in high-consequence environments already knows this — aviation, medicine, the military. Law enforcement is no exception, and it never has been. What’s been missing isn’t the training philosophy. It’s been the capacity to deliver enough reps to make it stick, for every officer, not just the ones who get picked for the annual exercise.
That’s the door AI actually opens. Not a smarter camera. Not a faster report. The ability to give every officer in a department unlimited, realistic, zero-consequence practice at exactly the decisions that matter most — de-escalation, crisis intervention, split-second judgment calls — as many times as it takes for the correct response to become automatic.
It doesn’t replace the instructor. It doesn’t set the standard. It doesn’t take the accountability. Those stay human, and they should. What it does is remove the economic ceiling that has always limited how much rehearsal an agency could afford to give its people.
Agencies that grasp this are quietly restructuring how they think about training budgets. Agencies that don’t are still buying cameras and calling it progress.
If you lead training or sit in command staff meetings, I’d genuinely like to know: what’s the biggest thing standing between your department and giving every officer real reps this year? Time, money, instructor bandwidth, or something more cultural?
Reply and tell me. I read all of them.
More on the operational side of this, including the verification frameworks I mentioned this week → https://elevateyourai.training
