Jevons Paradox
Making AI cheaper creates more usage, not less.
Jevons Paradox (1865) states that when the cost of a resource falls, total consumption rises — often dramatically — as new use cases become economically viable.
In AI this plays out constantly: as token prices drop, organisations don't use the same amount of AI more cheaply; they find entirely new applications that were previously unaffordable. Claude Haiku 4.5 at $0.80/M tokens unlocks checks that teams would never have commissioned at $15/M.
Two practical implications:
- Cost governance — cheaper models don't reduce your AI bill; they expand the scope of what gets automated
- Second-order volume — 10× velocity doesn't produce 10× productivity; it produces 10× output that must be reviewed, tested, and maintained by the same team
Surfaced in Google's DORA research and Adam Bender's Google I/O 2026 talk on Software Ecology.
In plain terms
When motorways improved, people didn't drive the same route faster — they drove 5× more miles. Fuel efficiency made us drive more, not less.