In 2021 a single publishable commercial image cost somewhere between three hundred and three thousand dollars, and required between one and forty-eight hours of human labour. By 2026 the same image, produced by the dominant generative platforms at near-final commercial quality, costs cents and takes seconds.
The same collapse has run through drafting, prototyping, rendering, scoring, editing and translation. It is not uniform and it is not complete, but it is far enough along to be an industrial fact rather than a research direction. That fact is the proximate cause of the imagination economy.
What collapsed, precisely
Not the cost of expertise. Not the cost of judgement. Not the cost of distribution in every channel. What collapsed is the marginal cost of a competent artifact — a draft that is good enough to evaluate, ship, or iterate on. The first unit and the thousandth unit now cost approximately the same, and both are near zero relative to the previous decade.
This is why volume strategies that once signalled capability — more designers, more writers, more render farms — no longer explain who wins. Volume of competent output is abundant. Quality of direction is not.
Why the official statistics miss it
Economic measurement still records the price of a software subscription, a designer's hour, a copywriter's wage. Those prices are roughly stable. The thing they were a means to — a finished artifact — has fallen by orders of magnitude.
The precedent is William Nordhaus on artificial light: across two centuries the price of a candle fell modestly, while the price of an hour of light fell by a factor of forty thousand. The Industrial Revolution did not make candles cheaper. It made the thing candles were for almost free. Generative systems are doing the same to artifacts. Wages and seat licenses are the candles.
What the collapse does not mean
It does not mean every domain is finished, or that taste is automated, or that organisations can fire their makers tomorrow. Long-horizon work, regulated work, and work whose failure modes are catastrophic still bind on execution. The claim is narrower: for a large and growing share of knowledge and creative production, the step that used to limit the system no longer does — so the binding constraint has moved upstream to specification.
How to recognise it in practice
Teams report the same pattern: they can generate more options than they can evaluate; meetings fill with near-final drafts that still feel wrong; the scarce calendar item is no longer “time to make” but “time to decide.” When that pattern appears, the cost collapse has already arrived for that workflow, whether or not the organisation has named it.