In December 1981, New Order's drummer Stephen Morris was sitting in a Manchester studio watching a strange light change the color of the room. "The light seemed to me to take on a crystalline glow, giving the room an aquamarine haze," he said. "Everything's gone green," he observed, out loud, to nobody in particular. Producer Martin Hannett liked it enough to name the track after it, then walked out of the mix over an argument about the drums and never worked with the band again. Nobody in that room thought they were describing enterprise AI reporting forty-five years early. But they were.
An employee described what AI adoption looks like once usage becomes the target: "They monitor our usage and pull people into meetings asking why they aren't using enough Claude credits."
Sit with that for a second. A leader looks at a dashboard, sees a low number next to someone's name, and books a meeting to ask why they aren't using enough AI. Not whether the work got faster. Not whether quality improved. Not whether rework dropped, or the client got a better result. Just, why aren't you consuming enough credits. A room goes a certain color, somebody names the color out loud, and everyone in the meeting treats the color as information. Nobody in that Manchester studio checked whether the green light meant the mix was actually good. Nobody in your leadership meeting is checking whether the green dashboard means the work actually got better either.
Once people know that's what leadership is actually measuring, the rest is entirely predictable. Leadership needs proof the rollout is working. Usage is the easiest number to measure. Usage becomes the target. Employees find ways to increase usage. The dashboard goes green. The work stays exactly the same.
The lesson isn't "don't measure AI adoption"
It's that the metric you choose becomes the behavior you get. Measure tokens, and you'll get more tokens. Measure logins, and you'll get more logins. Measure prompts, and you'll get more prompts. None of those tell you anything about whether the actual work changed. New Order found this out the hard way with the actual best-selling 12-inch single ever pressed. "Blue Monday" moved over a million copies in the UK alone, and Factory Records lost money on every single one of them, because nobody costed out the elaborate die-cut sleeve before they pressed it by the truckload. The chart position was real. The sales number was real. Neither one told anyone in the building whether the label was making money, and by the time somebody checked, the answer had been sitting there the whole time, in a much less exciting sentence than "best-selling single in history."
But measure the cycle time of one specific workflow, before and after AI, and now you're measuring something real. A lot of what gets labeled "AI resistance" inside a company may simply be employees responding rationally to what leadership chose to count. People aren't dumb. They optimize for the scoreboard that's actually in front of them, and if the scoreboard says "tokens," you will get tokens, whether or not a single piece of work got better because of it.
Before you schedule another training session to fix "adoption," look at the dashboard first. You might be measuring the wrong thing entirely, and no amount of enthusiasm training fixes a metric that's rewarding the wrong behavior by design.
If your AI dashboard is green but you have a quiet feeling the work hasn't actually changed, that feeling is probably correct. New Order never meant "Everything's Gone Green" as a warning about anything — it was a throwaway line about a weird light in a room. Forty-five years later it's still the most accurate description going of what a vanity dashboard actually is: something changed color, somebody named it, and the naming got mistaken for the truth. Tell me what you're measuring, and I'll tell you what it's likely hiding.