Cheaper Work Costs More
Cheaper inputs do not lower the bill. They raise the volume. The ceiling on AI is not its price. It is the judgment to use it.
AI is not a cost-cutting play.
Earlier in this body of work, the argument was that AI is electricity. Electricity is an input, and cheaper inputs do not get banked. They get wired into more rooms. The old pitch, that AI cuts costs, has been corrected. The correction has since overshot into a new mistake: treating rising AI spend as proof that the investment is working.
Cheaper inputs are not lower bills
Add lanes to a congested highway and the lanes fill. Driving gets cheaper per trip, so people take more trips, and the congestion returns at a larger scale. Economists call it induced demand. The older, formal version is Jevons and coal: as engines burned it more efficiently, the world burned more of it.
Cognition is the input that just got cheap. Drafting, summarizing, analyzing, comparing, rewriting. The price of a unit falls, the number of units rises, and total spend holds flat or climbs. Most informed people now agree on this. It is not the point of this article. It is the ground the point stands on.
Why the value holds
Consider what actually changed. Work that once took hours of skilled effort now runs on electricity. The unit of work is the same unit.
The work still has to be worth doing. A report nobody reads is cheaper to produce and worth exactly what it was before. AI changes what production costs and how the work gets done. It does not change what a unit of work is worth. Value per unit holds. The input is relocated and multiplied. The money moves to a different column, and then the column grows.
Spend is not a scoreboard
“AI cuts costs” is dead. The bills killed it.
The bullish reflex that replaced it is the new lazy framing: spend is rising, so we are winning. That reads like a scoreboard. It is not one. Induced demand is the default. It arrives whether the extra volume is insight or noise. A team that doubles its output of analysis has doubled its consumption. It has not necessarily doubled anything of value.
Rising spend tells you the tap is open. It says nothing about what is in the glass.
The ceiling is judgment
Cost is falling and will keep falling. What does not fall with it is the human capacity to read, check, and decide.
Production scales with tokens. Verification scales with attention. More drafts, more analyses, more options, all produced faster than anyone can responsibly review them. An earlier piece argued that accountability does not scale with autonomy. This is the throughput version of the same point. The binding constraint has moved from the price of producing to the capacity to verify and decide.
Every added draft is a claim on someone’s attention. Cheap production makes the claims arrive faster, and attention is the one input that has not gotten cheaper. When review falls behind, one of two things happens. Output ships unchecked, or output piles up unread. Both look like productivity on a dashboard. Neither is.
Output that no one has verified is not an asset. It is inventory with a liability attached.
Govern the volume
Whether to adopt is settled. Induced demand settled it. The job now is deciding what to do more of, and making sure verification can hold the new throughput.
Start with visibility. Cloud spending taught the playbook: tag the usage, assign an owner, and watch what the volume produces. Then ask one question of every stream of new output. Does it compound value, or does it burn attention?
Then decide what not to produce. When a unit of work costs almost nothing, the discipline that used to come free from the price tag disappears. Someone has to supply it on purpose.
Fund the review capacity as deliberately as the production. A workflow that doubles output without adding the means to verify it has not doubled productivity. It has doubled exposure.
The price of the work is falling. The volume is exploding. Leadership is no longer about buying the capability. It is about aiming it.