In this article
I've spent twenty years watching what people paid to write, draw, translate, code the routine stuff, and I'd never seen a price fall this fast. The rule operating behind it wasn't invented by AI or by 2026: oversupply has sunk prices for as long as there have been markets. When a good stops being scarce, it drops. What's new is the speed. What sold for fifty euros five years ago today comes out of a machine at zero cost, and what's left standing is what the machine doesn't make: being there, in person, signing with the body. Value migrates towards presence, and the migration leaves casualties. A lot of people in the trade find themselves mid-journey, with no floor under their feet and no place to hold out.
The basic rule and its speed
That the price falls when supply rises surprises no one. The novelty is the pace. A human produces, with luck, a few pieces of medium content a day: an article, an illustration, a batch of photos. A generative model produces thousands in the same time, and it does it for millions of people at once. Added up, the supply of reproducible content has multiplied by several orders of magnitude in barely four years, between 2022 and 2026.
The problem isn't only how much supply has risen. It's that the market is slow to react. There were trades and platforms that took decades to consolidate —stock photography, microcontent, derivative illustration— and now reorganise in one or two years. The economy runs; people don't. Whoever lived off those markets doesn't have the institutional time it takes to retrain without it costing them their salary.
What has already fallen
It's worth being cautious with the figures, because here imprecision has consequences and because a good part of what circulates are sector estimates, not audited balance sheets.
The low segment of stock photography is the clearest example. Generic illustration, the simple logo, the filler mock-up, all that now competes against a machine that delivers it free, and orders of that kind on the freelance platforms shift towards jobs that include something more than the image: direction, judgement, a demonstrable originality. You don't need to pin financial figures on Getty or Shutterstock that they haven't made public to see the pattern; it's enough to look at what gets paid for and what no longer does.
Something similar happens with technical translation. The total volume of translated text is sky-high, because machine translators are baked into any content manager, but the value left to the human translator has retreated to revision and specialisation. And with generic copywriting —product descriptions, search-oriented posts, bargain-bin newsletters— the same happens: demand for the writer who delivers correct, anonymous text sinks, while that for the one with a recognisable voice holds.
The case of routine code deserves a qualification the easy discourse usually skips. It's true that programming assistants accelerate repetitive tasks, and the study GitHub published in 2022 on Copilot measured that developers completed a specific coding task around 55% faster with the tool. But that 55% is from a bounded experiment, not a law of the trade, and it's best not to confuse it with the work of Brynjolfsson, Li and Raymond, who measured something else: customer-support agents, not programmers. The aggregate effect on junior-programmer employment is real, if diffuse: fewer routine posts per team.
The direction points entirely the same way. The floor of the content market is being left without economic value.
What rises
The symmetrical face of the phenomenon is the one almost nobody tells with the same detail. If the abundant drops, the scarce rises, and the scarce now is the present body.
Concerts and live events grow above the economy around them, and the price of a ticket to a medium live show has risen visibly to anyone who's tried to buy one. In-person conferences recovered after the pandemic and are filling halls again precisely in the sectors that produce the most digital content: tech, education, marketing. Courses with mandatory attendance —bootcamps, executive programmes— charge a clear premium over their on-screen equivalent, and not for the subject, which is the same, but for the room.
Subscriptions to creators with a voice of their own go the same way. The paid-newsletter platforms reward the identifiable author, not the interchangeable text; the gross income those platforms distribute to their authors is already counted in the hundreds of millions of dollars a year. And the services that are only worth it if there's a human across from you —therapy, weighty legal advice, consulting, accompaniment— hold or grow, not because AI can't imitate them, but because what's being bought is the certified person on the other side.
Economic value migrates towards what the machine doesn't produce, or doesn't produce in a way you can't tell apart. The in-person. The event. The human with a guarantee.
Pine and Gilmore, thirty years early
Joseph Pine and James Gilmore published The Experience Economy in 1999. They held that economic value had shifted across history from raw materials to products, from there to services and from there to experiences: the moment lived, what's remembered, what can't be packaged. At the time it sounded like consultant's exaggeration. Today it reads as description.
What they couldn't anticipate was the cause that would accelerate it. Their framework didn't reckon with generative AI because it didn't exist, and yet it fits without effort: any technology that cheapens the reproducible makes the irreproducible, by contrast, dearer. AI has done that work in five years. It has condensed into a single half-decade what the theory spread across generations.
The human signature as certificate
A new label has appeared that until recently would have sounded absurd: "made by human", "no AI", "100% human". Marking something as human was as redundant as marking water as wet. Now it distinguishes, because the opposite has become the default state.
There are artists who sell dearer precisely for declaring they don't use AI. Publishers who turn "no AI in our process" into a selling point. Newsletters and blogs that promise auditable human authorship. Writing services that guarantee there's no model behind them. The paradox holds on its own: what was the norm becomes the premium the moment it can no longer be taken for granted. The "written by human" seal of 2026 works like the "handmade" of the seventies, when the factory already produced everything and the artisanal became a declared luxury. From obvious to distinctive in a single decade.
The vanishing middle class
The hardest part of this whole story is in the middle. The bulk of content professionals were neither the bargain-bin generic creator nor the star with a personal brand. They were the majority who deliver good work without being exceptional, with a stable clientele and a viable working life. That band is the one being left without ground.
They're squeezed from both sides at once. From below, the machine delivers content good enough for many clients at zero cost, and the client who was satisfied with good-not-exceptional now settles for good-not-exceptional automatic and much cheaper. From above, the creator with a recognisable brand pulls away: the more the interchangeable abounds, the more whoever isn't interchangeable stands out.
There's no continuity between those two extremes, there's a gap. The middle band moves or empties out. Whoever has a personal brand rises; whoever doesn't drops to compete against AI on terrain where they already start losing. Economic sociology has christened this winner-take-most, as against the winner-take-all of old: not one alone wins, but the long tail shrinks and the prize concentrates in a narrow upper layer.
The transition without a net
The hardest part of this change is neither technical nor economic. It's institutional. Those crossing it now —junior designers, technical translators, filler illustrators, bargain-bin writers, programmers starting out— do so without the net other labour reconversions had.
The industrial transition of the nineteenth century was a butchery, but it ended up giving birth to answers: unions, social security, labour regulation. The digital one of the twentieth century produced an intermediate response, with vocational training and retraining subsidies. The AI transition, so far, has produced nothing equivalent. Each professional crosses it on their own.
This last is a personal reading, though defensible with data: the mismatch between the speed of the change, the magnitude of the blow and the slowness of the collective response is accumulating a social cost that at some point will turn into a political demand. When, I don't know. That it'll happen is made plausible by the history of every prior reconversion, which took their time but arrived.
Suspicion as a habit
In daily life this already shows, and not in the abstract. Faced with a text, an image or a video, the first question has stopped being "what does it say?" and has become "who made it?". What carries no signature is assumed to be machine. Authorship was the invisible frame within which we read; now it's a datum you have to go and look for.
Hence people pay for presence: for the concert, for the in-person medical consultation, for face-to-face therapy, for the dinner the chef comes out to sign. And hence the digital without a recognisable author is consumed by the bucketload and paid for ever less. The attention economy absorbs it without trouble; the value economy doesn't.
That whole equation is being rewritten in silence. With no political framework to name it, no public accounting to measure it, no educational programme to teach it. Whoever crosses the transition does so with a compass they've had to build themselves, and whoever can't read it stays in the gap left by the middle class.
Definitions
Zero marginal cost: a characteristic of digital goods —and of AI-generated content in particular— by which producing one more unit costs essentially nothing, which drives up supply and sinks the price.
Winner-take-most: a market dynamic in which the prize concentrates in a narrow upper layer and the middle band contracts. A modern variant of the classic winner-take-all, in which a single winner takes it all.
Market bifurcation: the separation between a low segment, turned into an indistinguishable commodity, and a high premium segment, with a hollowing-out of the middle band.
Verified human / made by human: an emerging commercial label that certifies a piece of content or a service was produced by a person without significant AI assistance.
Experience economy: the framework proposed by Pine and Gilmore that places lived experience as the economic category of highest added value, at the end of a historical progression starting in raw materials.
References
Pine, B. J. and Gilmore, J. H., The Experience Economy (Harvard Business School Press, 1999), origin of the thesis on the historical shift of value towards experience, which structures the article.
GitHub, "Research: quantifying GitHub Copilot's impact on developer productivity and happiness" (2022), the study the figure of ~55% faster on a coding task with an assistant comes from. Available at github.blog/2022-09-07-research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness/
Brynjolfsson, E., Li, D. and Raymond, L., "Generative AI at Work", NBER Working Paper 31161 (2023), a study on customer-support agents —not on programming—, cited here to undo a usual attribution. Available at nber.org/papers/w31161
Sacra, financial profile of Substack, and Backlinko, "Substack users", sources of the figure of gross income to authors in the hundreds of millions of dollars a year (≈370 million in 2024, ≈450 million in 2025). Available at sacra.com/c/substack/ and backlinko.com/substack-users
Anderson, C., Free: The Future of a Radical Price (Hyperion, 2009), on the economics of the zero price in digital goods.
Susskind, D., A World Without Work (Metropolitan Books, 2020), on the labour impact of advanced automation.
Further reading
Yochai Benkler, The Wealth of Networks (Yale University Press, 2006), a framework on how collaborative production of digital goods reorganises the cultural economy; useful for not reading the current devaluation as decadence but as reorganisation with different winners and losers.
Astra Taylor, The People's Platform (Metropolitan Books, 2014), an anticipatory critique of how digital platforms promise democratisation and produce concentration; it applies directly to the content economy in the AI era.
Tim Wu, The Attention Merchants (Knopf, 2016), a history of the attention economy that serves as a backdrop for understanding the migration of value towards presence as the continuation of a long-running process, not as a novelty induced by AI.
You may also be interested in
- Authenticity as a luxury. The middle class of hybrid content nobody knows how to value
- Work as residue. The frontier moves every year
- The collapse of the price of knowledge
- The entropy of digital content
Comments0
No comments yet.
Leave a comment