Most organizations I talk to have already bought the AI. The licenses are paid. The Slack channel exists. Someone, somewhere, ran the workshop. And yet a quarter later, usage flatlines around the same three early adopters who would have figured it out anyway.
The reflex is to blame culture, training, or change management. I think it’s simpler and more mechanical than that. AI adoption inside an organization is a habit problem, and habits have a known anatomy: cue, craving, response, reward. If any of the four is missing, the loop never closes — and a loop that doesn’t close never becomes automatic.
Cue. A habit needs a trigger that lives in the existing workflow. “Use AI more” is not a cue. “Before sending any client email longer than three paragraphs, paste it into Claude” is a cue. Most rollouts skip this entirely and hope the tool itself becomes the cue, which is like hoping a gym membership becomes a workout.
Craving. People don’t crave AI. They crave the outcome — the meeting prep that doesn’t eat their Sunday, the analysis that used to take a day. If the team can’t articulate, in their own words, what specific pain the tool removes for them, there is no craving. There’s just a tab open in the browser.
Response. This is where most rollouts quietly die. The response — the actual act of using the tool — has to be lower-friction than the alternative. If opening Claude, framing the prompt, and iterating takes longer than just doing the thing the old way, the brain correctly routes around it. Every extra click is a tax on adoption. Most “AI strategies” I see ignore friction entirely and substitute enthusiasm.
Reward. And then: did it actually work? Was the output good enough to ship, or did it need so much editing that the person silently concludes it was faster without? The reward has to be felt, immediately and repeatedly, in the same session. Delayed or ambiguous rewards do not build habits. They build skepticism.
Organizations are habit machines made of habit machines. An individual loop closing is not enough. The team’s rituals — the standup, the review, the weekly report — are the cues for the next layer up. Which means the question isn’t “are people using AI” but “which of our existing rituals now have an AI step inside them, and which don’t.” Everything else is theatre.
This is what I think most of the AI-transformation conversation gets wrong. It treats adoption as a decision, when it’s a redesign. You don’t decide to have a habit. You build the loop, or you don’t.
The honest version of the question is uncomfortable: which three workflows in your week are you willing to re-engineer down to the cue? Not augment. Re-engineer. If the answer is none, the licenses are decoration.