The Jevons paradox says that using a resource more efficiently can raise its total consumption instead of lowering it. It happens when use grows more than the savings: if something goes further and gets cheaper, people use it for many more things. William Stanley Jevons described it in 1865, with coal.
A question from the app. Pick an answer.
Each request now costs half, and you now make four times as many. Your monthly bill?
It would halve if you used it the same. With four times the requests at half the price, you pay twice as much as before.
Half the price times four times the use: 0.5 × 4 = 2. Getting cheaper invited more use, and spending went up: the Jevons paradox.
They would cancel out if usage doubled. With four times the use, 0.5 × 4 = 2: the bill doubles.
Intuition says that if you learn to do the same work with less coal, you will burn less coal. In 1865 Jevons warned the opposite could happen: if using coal pays off more, it makes sense to use it in more places, and total consumption goes up.
It is not a law that always holds. If you do the same work with less of a resource and use does not grow, you consume less. Salt costing half as much does not make you put twice as much on your food: you use the same salt and spend less.
The paradox occurs when improved efficiency drives enough extra use for total resource consumption to exceed its previous level.
The same can happen with AI. If each request to a model costs less, it becomes reasonable to use it for things that were not worth it before: checking every ticket, every message, every form. The cost per request goes down, but total use and total spending can go up. Can, not must.
The practical lesson: when you lower the cost per request, watch the volume too. A tiny rate times millions of requests stops being tiny.
Diogo Almeida, CEO of TypeSafe AI, said on X that Jev is named after the Jevons paradox. The bet behind the name: that cheaper intelligence will be used far more. It is a bet, not a guarantee.