There is a strange assumption buried in much of the anxiety surrounding generative AI: if machines can produce music, images, prose, video, and other creative works cheaply and at enormous scale, the market for human creativity must eventually disappear.

I suspect something very different is going to happen.

AI doesn’t necessarily destroy the market for human creativity. It could commoditize ordinary creative production while simultaneously increasing the premium attached to demonstrably human craftsmanship.

If that happens, we may eventually look back on our present arguments about AI art and recognize them as the early stages of a market sorting itself out. Rather than AI replacing human creativity wholesale, we may be watching two increasingly distinct markets emerge: one built around abundance, speed, personalization, and efficiency, and another built around scarcity, craftsmanship, provenance, and human performance.

There is nothing particularly strange about this. Markets have done it before.

Consider the automobile. A modern mass-produced car is an extraordinary technological achievement. Manufacturing automation allows companies to produce vehicles with a consistency, speed, and affordability that would have been unimaginable when automobiles were individually constructed. If transportation is the objective, mass production has been an overwhelming success.

Yet mass production did not eliminate the market for automobiles involving extensive hand craftsmanship. In some ways, it helped define it.

A Bentley isn’t valuable simply because it can transport someone from Detroit to Chicago. A Ford can do that perfectly well. In fact, depending upon what one values in an automobile, the Ford may perform the basic transportation function just as reliably and at a fraction of the cost.

But that isn’t entirely what the Bentley customer is purchasing.

Part of the value proposition is precisely what mass production attempts to minimize: human attention, craftsmanship, scarcity, materials, time, and labor. The inefficiency is not necessarily a defect. In the luxury market, some of that inefficiency becomes part of the product.

We see the same phenomenon everywhere. Factory-made furniture exists alongside handmade furniture. Quartz watches, which can keep extraordinarily accurate time, exist alongside expensive mechanical watches. Mass-produced clothing exists alongside bespoke tailoring. Digital prints exist alongside original paintings.

Technology makes the functional product more abundant and inexpensive, but abundance doesn’t necessarily destroy craftsmanship. Instead, it separates craftsmanship from utility.

That distinction may become increasingly important in the creative arts because generative AI is doing something extraordinary to the economics of creative production: it is removing friction.

We normally think of that as unquestionably good. Waiting is bad. Expense is bad. Difficulty is bad. Limited production is bad. Automation creates economic value precisely because it removes those constraints.

But friction has another characteristic that is easy to overlook.

Sometimes friction is evidence of craftsmanship.

It took someone years to learn how to play that instrument. It took weeks to paint that canvas. Someone carved that table by hand. Someone spent years learning the mechanics of storytelling and then months or years writing and revising a novel.

None of those things necessarily proves that the resulting product is objectively better. A handmade chair can be uncomfortable. A traditionally painted portrait can be ugly. A novel that took ten years to write can still be terrible.

But the friction tells us something about what was required to produce the thing.

That matters much more when technology makes the alternative nearly effortless.

If I can generate 100 attractive images this afternoon, the existence of an attractive image is no longer particularly scarce. If I can generate dozens of competent songs over a weekend, producing a polished recording becomes less remarkable as an economic event. If enormous numbers of coherent novels can eventually be generated rapidly, simply producing 80,000 readable words will no longer tell us very much.

The supply increases, and markets generally respond to abundance by reducing the value attached to what has become abundant.

But not everything associated with creative production becomes abundant.

Human time doesn’t.

Human skill doesn’t.

Human performance doesn’t.

Human craftsmanship doesn’t.

And demonstrable human provenance doesn’t.

This is where our current moment begins to look surprisingly familiar.

The technology is new, but the social cycle surrounding it is not. Transformative technologies have repeatedly moved through some variation of breakthrough, rapid adoption, perceived threat, backlash, attempts to establish boundaries, accommodation, market segmentation, and finally normalization.

Industrial machinery threatened skilled craftspeople whose economic value had been tied to things machines could suddenly produce more efficiently. Photography raised serious questions about what it meant to create an image when a machine could capture one. Recorded music threatened working musicians because a performance could suddenly be reproduced indefinitely without musicians being physically present. Digital photography, synthesizers, sampling, CGI, desktop publishing, and countless other technologies produced their own arguments about authenticity and legitimate craftsmanship.

Those concerns weren’t necessarily foolish. Technological transitions create genuine economic casualties. Jobs disappear. Skills that once commanded premiums lose value. Industries restructure. Laws lag behind technology. People who invested decades learning how to do something understandably become concerned when a machine suddenly performs portions of that task almost instantly.

But history suggests that the eventual outcome isn’t always the disappearance of the older form.

More often, the market reorganizes around the new abundance.

Machine-manufactured textiles became ordinary, while handmade textiles became something worth specifically identifying. Photography became ubiquitous, while painting remained valuable. Recorded music became the easiest and cheapest way to hear music, while live performance developed into a separate experience for which people willingly pay substantial premiums.

Nobody goes to a Taylor Swift concert because attending a stadium concert is the most efficient way to hear a Taylor Swift song. Streaming the recording at home is dramatically easier and cheaper.

The inefficiency is part of what the customer purchased.

And that observation helps locate where we probably are in the generative AI cycle.

The mass-market breakthrough arrived around 2022 and 2023. The first stage was astonishment: Look what this can do.

Then came rapid adoption.

Then, predictably, came backlash.

This isn’t real art.

This isn’t writing.

This isn’t music.

It should be labeled.

It shouldn’t qualify for awards.

Publishers shouldn’t accept it.

It’s taking people’s jobs.

It was trained on stolen work.

We now appear to be moving into the boundary-setting period. Courts are being asked to resolve copyright questions. Publishers and creative organizations are developing policies. Platforms are experimenting with disclosure requirements. Creators are debating acceptable and unacceptable uses. Entire arguments are emerging around distinctions such as assistive AI versus generative AI.

If history is any guide, this period may last years. Earlier technological disruptions sometimes took decades before their economic and cultural consequences became ordinary. There is no reason to assume generative AI will settle immediately.

But the eventual destination may be easier to anticipate than the timetable.

At some point, people stop arguing constantly about whether the technology should exist and begin deciding which market they want to participate in.

That is where I suspect the creative economy is headed.

One market will be primarily concerned with output. The customer wants a result. They need background music for a video, an illustration for an advertisement, copy for a product page, an audio book voice, an entertaining story, or personalized media. If AI can provide an excellent result in seconds for pennies, efficiency wins.

There is nothing irrational or morally suspicious about that. Most people don’t demand that every chair they purchase be handmade. Sometimes they simply want somewhere to sit.

But alongside that market, another market can develop around provenance.

Who made this?

How was it made?

What did the person actually do?

Can the musician actually perform this song?

Can the painter actually paint?

Can the illustrator actually draw?

Did the writer actually exercise authorship over this book?

Suddenly, the human process becomes part of the product.

Imagine two equally beautiful recordings. One was generated through an AI music system. The other was written and recorded by musicians who can walk onto a stage and perform it.

As audio files, both might satisfy the listener equally well. But they aren’t necessarily the same kind of product.

One is an artifact.

The other is an artifact plus demonstrated human capability.

The audience can buy a ticket and watch the second one happen.

Visual artists may encounter the same phenomenon. Watching an artist actually paint may become more culturally significant when millions of aesthetically pleasing images can be generated instantly. The artist isn’t merely selling an image. The artist possesses and demonstrates a craft.

Writers may eventually confront their own version of that distinction. The question won’t necessarily remain merely, Is this a good book? It may increasingly become, What relationship does this person actually have to the creation of this book?

Publishing itself already gives us a smaller version of this phenomenon.

Traditional publishing contains enormous amounts of friction. An author may spend years writing a manuscript and then face querying, agents, submissions, acquisitions, editing, revision, production schedules, distribution, and finally publication.

Self-publishing removed many of those barriers, which has been enormously beneficial. Talented writers no longer need permission from an institutional gatekeeper to reach readers.

But removing the gate also removed one of the signals created by the gate.

A traditionally published book implicitly communicates that someone other than the author decided the work was worth publishing. That doesn’t mean the book is better. It certainly doesn’t mean a self-published book is inferior. It simply means the traditional publishing process contains friction, and surviving that friction communicates information to the marketplace.

Generative AI introduces the same phenomenon at a much deeper level. We aren’t merely removing friction from publishing creative work. We are removing enormous amounts of friction from producing creative work in the first place.

And when friction disappears from one part of a market, its presence elsewhere becomes more noticeable.

This produces the great paradox of AI abundance. The better AI becomes at producing creative artifacts, the less remarkable the mere production of those artifacts becomes. At precisely the same time, however, demonstrating human craftsmanship may become more remarkable.

AI may therefore decrease the scarcity of creative products while increasing the scarcity of demonstrably human creative processes.

That is why I don’t think the future necessarily looks like AI art fighting human art until one side wins. It may look much more like mass production and craftsmanship occupying different portions of the same marketplace.

Ford did not eliminate Bentley. IKEA did not eliminate handcrafted furniture. Photography did not eliminate painting. Recorded music did not eliminate live performance. Digital photography did not eliminate film.

In many cases, technological efficiency changed why consumers valued the less-efficient alternative.

And that leads me to wonder whether we’re currently trying to label the wrong thing.

Most of today’s conversation assumes human creation is the default and AI creation is the exception. Therefore, the proposed solution is usually to label the exception: AI-generated.

That makes sense while AI-generated material remains unusual.

But what happens if it becomes ubiquitous?

If AI-generated and AI-assisted material becomes an ordinary part of the information environment, putting an AI label on everything could eventually become something like putting a machine-manufactured sticker on almost everything in a department store.

That isn’t generally how mature markets handle the distinction.

We don’t usually label furniture machine-made. We label certain furniture handcrafted.

We don’t label ordinary clothing mass-produced. We identify certain clothing as handmade or bespoke.

We don’t warn consumers that most of the music they hear has been technologically recorded and processed. But live performance means something.

Perhaps creative markets eventually undergo the same reversal.

Instead of attempting to identify every artifact involving artificial intelligence, perhaps the economically meaningful credential becomes human authored, human performed, human crafted, or some stronger form of verified human creation.

That changes the question considerably.

Instead of requiring every creator using AI to prove that something synthetic occurred, creators who want the premium associated with human craftsmanship would demonstrate that their work has the human provenance they claim.

Everything else could simply remain uncertified.

This approach may also make more sense technologically. As synthetic output improves, detecting AI from the finished artifact is likely to become increasingly unreliable. If an AI-generated song sounds indistinguishable from a human recording, inspecting the finished audio may tell us very little about how it came into existence. The same problem applies to images and increasingly to prose.

Perhaps we are approaching the problem backward.

Rather than staring at a finished artifact and asking, Was AI involved?, the more durable question may become, Can the claimed human creative process be demonstrated?

That moves us away from detection and toward provenance.

And importantly, none of this requires us to declare AI-assisted work illegitimate. Nor does it require everything to be handmade. Different production methods can simply produce different categories of products for different markets.

One customer may want the best song available for the lowest price. Another may specifically want music written and performed by musicians.

One customer may want an attractive image. Another may want a painting created by a particular artist.

One reader may simply want an entertaining story. Another may care deeply that a human author actually conceived, developed, wrote, and revised it.

Neither market has to destroy the other.

They can coexist because they are selling different things.

This is ultimately an old economic story wearing new technological clothing. Industrialization didn’t eliminate craftsmanship. It changed its meaning. Once almost everything could be manufactured efficiently, handmade became valuable information about an object’s provenance.

Generative AI may do something very similar to creative work.

When creative artifacts were difficult to produce, we could largely take the human process behind them for granted. When creative artifacts become virtually unlimited, we won’t be able to.

And perhaps that tells us something about what comes next.

The abundant product doesn’t particularly need to explain why it is abundant. The scarce product needs to establish why it deserves to be treated as scarce.

The age of AI may therefore not make human craftsmanship obsolete. It may force us to recognize that craftsmanship itself was part of what we were valuing all along.

And if that is where the market is heading, perhaps the future of creative labeling can be summarized very simply:

Abundance doesn’t need a certificate. Scarcity does.