Most organizations measuring AI success are measuring adoption, not value. Usage dashboards, seat counts, prompt volume - these are the metrics you track when you cannot yet define what "working" looks like. The activity looks like progress. It isn't.
This is not unique to AI. Every major technology adoption cycle produces a version of this problem: when you can't measure the outcome, you measure the input and call it a proxy. Cloud adoption was counted in workloads migrated. Digital transformation was counted in lines shipped. Agile transitions were counted in story points. AI is being counted in tokens consumed.
The difference is that the token bill arrives monthly and is not small. And it will grow quickly. Tokenmaxxing will be extremely expensive once the major AI platforms go public.
Measuring AI adoption by usage is like measuring the health of a business by how many meetings its employees attend. The metric is real. The correlation to what you actually care about is weak. And optimizing it produces exactly the behavior you would expect: people using AI in ways that are visible, reportable, and defensible - regardless of whether it changes anything that matters.
I'm a heavy AI user myself. I practically live in Claude Code. But I couldn't care less about token KPIs - not my own, let alone anyone else's token exhaust. The people and organizations getting real value from AI are not the ones with the highest usage rates. They are the ones who decided what "value" meant before they bought a single license.
Thoughts? Find me on Bluesky.