Your company now runs on two currencies
One of them is collapsing in price. The other never does. And there is no exchange rate between them — which is why nobody can say whether the account closes.
Picture a company where half the purchases are paid in one currency and the other half in a currency nobody knows how to convert. Each area has its own, each supplier charges in theirs, and there is no conversion table anywhere. You can work like that for a while — until someone asks what the month cost.
That is where adopting AI has put almost every company, and nobody announced the change.
The two currencies
The world is reducing work to two costs: Time and Token.
Token is the cost of artificial cognition — the price of one unit of machine reasoning. It is collapsing: cheaper with every generation of model.
Time is the cost of human cognition — the price of judgement, of attention, of the decision. It does not collapse: it is the resource that becomes scarce.
Whoever looks only at the first concludes that AI got cheap. And it did, per unit. The trouble is that nobody stops at the unit.
Why the bill rises when the price falls
Getting cheaper is not the same as saving.
What used to be asked sparingly of a specialist starts being asked at any hour, by anyone, about anything. The price per question collapsed; the number of questions multiplied. It got cheap per unit and expensive in total — and nobody knows how to measure the total.
Look at what your company has today. Probably several models in use: an assistant here, another embedded in some tool there, one more that somebody subscribed to alone. Each with its own price, its own meter and its own unit. No common unit between them.
And then the simple question has no answer: how much artificial cognition did the company consume last month? Requested by whom? Under what rule? And how much human time did that free up — or demand back, in checking, correcting and redoing what came out wrong wearing the face of right?
The half nobody adds up
That last part is the one that usually disappears from the account.
Every generated answer goes back to a person who has to read it, judge whether it is right, and own the result. That time is real, it is expensive, and it is precisely the resource that did not get cheaper. When it stays out of the account, AI looks like pure gain — and what is being measured is half an operation.
This is not an argument against using AI. It is an argument against using it without a ruler: what is not measured is not governed, and with no unit of account there is no governance — only the feeling that it is paying off.
How to tell whether your company has the ruler
One question is enough, and it is not a technical one:
Is there anywhere you can see, side by side, how much the company spent on AI this month and how much human work that moved?
If the answer is a supplier's invoice, that is not the ruler — it is the statement of one side. If it is a spreadsheet someone updates when they remember, it is not either: a ruler that depends on someone remembering is not a ruler, it is an intention.
And if there is no answer at all, that is not carelessness. It is the normal state of almost every company right now — what differs is how long each one takes to notice.
What to do before you have the ruler
Measuring everything is the big step. The small one comes first, and fits in an afternoon: know what is in use.
Which AI tools your team uses today, who pays for each one, and who decided it was allowed. Without that list, any number you produce later will measure only the part you already knew about — and what hurts is usually in the other part.