Jevons Dial: cheaper per task, dearer per year

A firm adopts an AI drafting tool: $1.00 a task, 100 tasks a day, $100 a day. Engineering keeps making it more efficient. Drag the cost down and watch the bill; then flip the toggle and see the one assumption the whole paradox hangs on.

The dial

efficiency vs launch

total spend per day
tasks per day
energy per day (kWh)

The demand curve · price vs tasks (log-log)

What unlocks as it gets cheap

Efficiency is a throttle on price, not on appetite. William Jevons watched better steam engines increase Britain's coal burn in 1865, and the mechanism has not changed: when a unit of work gets cheap enough, people find new work to do with it, and total consumption rises through the efficiency gain. Whether that happens to your AI budget hangs on one question the model cannot answer: what will your people start doing once it is this cheap? Every line item in the unlock list is a human decision. So are the emissions, and so is the headcount plan that assumed "more efficient" meant "less used".