The cost of producing one additional unit of AI reasoning or output, which has been falling steeply and keeps falling. For agencies, it is the economic force sitting underneath every pricing, scoping, and headcount conversation currently happening.
Also known as cost per token, cost of cognition, unit cost of AI output
The marginal cost of intelligence is what it costs to produce the next analysis, the next draft, the next hundred variants, once the system is built. It has collapsed on two fronts at once: model providers keep cutting per-token prices, and cheaper models keep reaching quality levels that used to require expensive ones.
The phrase is doing analytical work rather than describing a specific technology. It names the reason so many agency business assumptions stopped holding at the same time, because almost all of them were built on cognitive work being expensive and slow to produce.
Agencies have historically sold time, and time was a reasonable proxy for the cost of thinking. That proxy is breaking.
Hourly billing is the exposed model. If a deliverable that took twenty hours now takes three, honest hourly billing is a pay cut for getting better. The pressure to move toward value, output, or retainer-based pricing is not a fashion, it is arithmetic.
Scarcity moves upstream. When producing options is nearly free, the constraint becomes knowing which option is right and being willing to stand behind it. Judgment, taste, and accountability are the things that did not get cheaper.
Client expectations reset faster than contracts. Clients read the same trend pieces. The scope conversation is increasingly about why a thing costs what it costs when the client believes the machine did most of it, and agencies need a clear answer that is not defensive.
An agency reviews a retainer where a monthly content package was scoped at roughly ninety hours. After building an AI-assisted production workflow the same package takes closer to thirty, and the client has noticed the turnaround got faster. Rather than quietly banking the margin or waiting to be squeezed, the team restructures the retainer: a fixed fee for strategy and quarterly positioning work priced on the judgment involved, a per-asset rate for the production package priced well below the old hourly equivalent, and a new line for maintaining the workflow itself, including the evaluation suite and the brand checks that keep the output usable. Total contract value stays roughly flat, but nothing in it is priced on how long a person sat at a screen.
The automations and agents module of the workshop teaches you how to build AI workflows that compress the busywork without taking the craft out of the studio.