The new performance review question isn't "did the AI help?" It's "how many tokens did you burn?" Across a growing set of companies, adoption is being measured by consumption — seats logged in, prompts fired, credits spent — and the people being measured have adapted the way people always adapt to bad metrics. They perform the number.
Feigning success
Middle managers, caught between a mandate from above and a team that genuinely has no good use for the tool, are increasingly reporting AI wins they can't point to in the product. The incentive isn't to ship something useful. It's to show motion. A summarized doc nobody reads. A chatbot demo that never reaches a customer. Evidence of AI, manufactured for the quarterly deck.
Burning credits to hit quota
The sharper version of this is the "token Hunger Games" — workers told to prove productivity by burning through AI credits, whether or not the work needed them. Spend the allocation or look like you're behind. So they spend it. The dashboard lights up green. The roadmap doesn't move.
What the metric is actually saying
A usage metric only works when usage correlates with outcomes, and in the early, awkward phase of AI adoption it usually doesn't. Tracking consumption as if it were progress doesn't reveal where the tool helps. It rewards the teams best at looking busy and quietly punishes the ones who spent the quarter finding out the model wasn't the right call.
The fix isn't less measurement. It's measuring the thing after the model — the ticket closed, the bug found, the customer retained — instead of the tokens on the way in.
Get in touch if your adoption numbers look great and your output doesn't.