The Clone Problem

In nineteenth-century England, William Morris looked at what industrial mass production was doing and saw something rotten. Cheap furniture, textiles, wallpaper; work broken into repetitive tasks, the maker disconnected from the thing being made. The loss of pleasure in the work mattered to him as much as the ugliness of the result.¹ The objects could be functional. They could be affordable. And they could still be dead.

Morris became a central figure in the Arts and Crafts movement because he understood that something essential was being lost in the name of efficiency. The hand of the maker. The imperfections that proved something had been touched by a human who gave a damn. The aura of the thing, to borrow Walter Benjamin’s word² - the presence and history of an original that a reproduction cannot inherit simply by looking the same.

That was more than 150 years ago. The machines have gotten better, the problem has gotten worse, and now we’re feeding the output back into the input.

What Mastery Actually Looks Like

Around 1508, Leonardo da Vinci was with an old man in Santa Maria Nuova hospital in Florence. The man told him he was over a hundred years old. When he died, apparently at peace, Leonardo dissected him to find out why a body could stop living without an obvious struggle.³

This was part of years of anatomical work, perhaps thirty human dissections as well as animal studies. Leonardo stripped a shoulder down in stages and drew a hand layer by layer. He injected molten wax into the cavities of a brain to discover their shape. Then he drew what he found, using multiple and exploded views to make the machinery of life visible. He wanted to know what was underneath the skin, and the investigation took him far beyond what a painting required.³

There was no product roadmap that required a wax cast of a brain. It was an act of obsessive, unreasonable commitment to depth, driven by the conviction that surface-level understanding produces surface-level work.

Michelangelo spent four years painting the Sistine Chapel ceiling.⁴ Head craned backward, body contorted, paint dripping onto his face. He wrote a poem to his friend Giovanni about the experience:

…my breast-bone visibly
Grows like a harp: a rich embroidery
Bedews my face from brush-drops thick and thin.

Michelangelo, poem to Giovanni da Pistoia; translated by John Addington Symonds.⁵

The poem ends with him declaring that “painting is my shame” - which, coming from the man who painted the most famous ceiling in human history, tells you something about the relationship between mastery and self-assessment.

Before any of this came the bottega: the workshop. Leonardo trained under Verrocchio; Michelangelo learned painting under Ghirlandaio.⁶ An apprentice’s education included grinding pigments, preparing panels, gilding, and sweeping the floor.⁷ The materials were part of the education. Paint had to be made before it could be put on a surface. The system was slow and unglamorous, and it understood something about mastery that we’ve almost entirely forgotten: you can’t skip the grinding.

This is what depth looks like: not talent, not efficiency, not the right tools, but a willingness to go further into the work than anyone else is prepared to go, for longer than anyone else is prepared to endure, because you understand that the surface can only be as good as the foundation beneath it.

The Convergence Machine

Open ten AI-generated websites in a row. The same hero section. The same gradient. The same three-column feature grid. The same testimonial carousel. Swap the logos and nobody would notice. Open ten AI-generated blog posts. The same cadence. The same hedging. The same structure of “here’s a problem, here are three points, here’s a conclusion.” None of it is bad, exactly. All of it is the same.

You don’t need every model to be mathematically condemned to mediocrity for this to happen. You need a process that keeps rewarding it.

Ask for a professional website. Accept the familiar layout because it looks finished. Ask for a cleaner article. Accept the smoother sentences without checking what disappeared. Give the result to another agent for review, and let it sand down whatever the first one left. Each pass can look like an improvement while the work becomes less particular.

That experiment lets the systems run without a person intervening. We can intervene, and still make the same mistake. A model can offer a strange, interesting answer; the person reviewing it can ask for something safer. Or the person can supply the strange, interesting detail, and the model can quietly take it out. Calling the result “polished” doesn’t make it better.

Left to decide what deserves to be in the work, AI is the ultimate junior practitioner: technically proficient, expressively empty. It can produce the appearance of judgement at industrial scale. It has not spent your years learning why this particular choice matters. If you hand that choice over, the part of the work that was yours becomes optional.

The Dead Internet

You open a page, then another, then another. Different domains, different subjects, the same polished feeling that nobody had anything particular to say.

The Dead Internet Theory used to sound like a conspiracy. Spend enough time wading through generated content and you understand the feeling behind it. Pages that nobody asked for, nobody needs, and nobody can distinguish from the page next to them. A city where every building is a slightly different shade of beige.

People can feel the difference even when they cannot explain it. They might not be able to articulate exactly what’s wrong, but something in the hindbrain knows. There’s no weight to it, no risk, no sense that someone had to actually think before they put words on a page. It’s the uncanny valley applied not to faces but to judgement.

New startups launch every day, and too many look like a clone of whatever template is currently trending. It’s like sitting through a Hollywood movie - the formulaic predictability detracts from any substance that might be lurking beneath the surface. The difference now is scale. What used to be a human tendency toward imitation has been industrialised. Generate from the familiar, approve what looks familiar, feed it back in. The platonic clone. The clone of clones.

The Death Spiral

Here’s where it gets disturbing.

Once generated work becomes training material, the loss can reach into the next generation of models. In Nature, Shumailov and colleagues studied recursive training on model-generated data and found that rare parts of the original distribution could disappear first. With further generations, the learned distribution could lose its resemblance to the original. They called it model collapse.⁹

How the data is handled matters. A separate study found stable results when original data was retained and generated data accumulated alongside it, rather than replacing it.¹⁰ Keeping the originals changes what survives. That is a reason to protect the source, not an excuse to stop worrying about what the copies lose.

Leonardo’s anatomical drawings exist at the extreme edge of human expression. They’re the weird, obsessive, deeply personal work that only exists because a specific human being kept opening bodies and asking questions that a respectable painting career didn’t require. That is the kind of work I mean when I talk about the long tail. The analogy matters because the unusual detail is so easy to discard. You can keep the subject, the technique, even the look, and lose the obsession that made it worth encountering.

Michelangelo’s ceiling is unreproducible in that sense. Not because the technique is complex; technique can be studied and replicated. The conditions that produced it are unrepeatable: four years of physical strain, an artist working at the limit of his capabilities, a poem of despair written in the middle of it. The work contains the struggle. The aura is the struggle, made visible. A probability engine can generate something that looks similar. It cannot give that image the life that made the original what it is.

In code, the same habit leaves its own damage. The code works and the tests pass. The model has added an abstraction because it recognises the pattern, even though the project already has a place for that responsibility. Now two pieces of code own something that should have had one owner. Each looks reasonable on its own. Someone still has to notice what happened to the system.

“Vibe coding” they call it: accepting whatever the model gives you because it looks right and runs without errors, never questioning the structure or the architecture. It’s the clone problem applied to engineering itself. Leonardo opened bodies to understand what was beneath the surface. Vibe coding doesn’t even look.

You see the same thing beyond code. The generated output often works. It passes. It ships. What it doesn’t reliably carry is judgement. The abstraction is a little too generic. The boundary is in the wrong place. The naming is technically fine and spiritually dead. The system gets more code and less shape.

We rebuilt our own documentation tooling because the existing options kept converging on the same output: functional, acceptable, forgettable. Once the average is cheap, taste stops being decoration and becomes the whole game.

The Edit That Removes the Person

I asked for a language pass on my own writing and got back something smoother that had lost the stories. The details that made it mine had been replaced by cleaner summaries. I wasn’t asking for the typos to be preserved. I was asking for the writing to be improved.

Tea ceremonies with martial arts masters in Hong Kong became “meeting other masters.” Einstein’s childlike innocence became a general observation about insight. The replacements weren’t nonsense. That was what made them so bad. They were plausible sentences standing where the reason to read had been.

A story about travelling alone, having everything stolen, and loading trucks to survive is not interchangeable with a sentence about resilience. The summary has thrown away the experience the reader was meant to encounter. It has kept the category and deleted the life.

And I had explicitly asked for those edges to stay. More instructions didn’t fix the judgement being applied to the work. “Improve this” was still being interpreted as “make this resemble other finished writing.”

A useful edit should know why the tea is there. It should help me say what happened more clearly, or find where the thought goes further. If a date is wrong, correct the date. If I turn a sensation into a claim about anatomy, make the distinction clear. Neither gives you permission to take the room, the people, or my conviction out of the story.

There is nothing sacred about making every task laborious. I don’t want to grind pigments for the sake of suffering. I want the time a tool saves to go into the decisions that make the work worth doing. A familiar search box helps someone find the material. A familiar sentence that replaces the material has done the opposite.

The Fork

When everything can be polished to a mirror shine by a machine, perfection becomes suspicious. It tells you nothing about who made it or why. It could have been made by anyone, which means it was made by no one.

You know the difference when you feel it. You’re scrolling through generated slop - competent, frictionless, dead - and then you hit something that stops you. A sentence that risks too much. A design choice that shouldn’t work but does. Code architecture where someone clearly agonised over the right abstraction instead of accepting the first suggestion. The rough edge, the human choice, the thing that didn’t need to be there but is. Fingerprints. Proof of life.

That’s the fork. Not between using AI and not using AI - that’s a false choice, as pointless as rejecting the printing press. The fork is between letting the machine do your thinking and using the machine to amplify thinking you’ve already done; between accepting the most probable output and insisting on the right one; between grinding the pigments and skipping straight to the canvas.

In practice this means refusing first-draft architecture, reworking the copy until it carries a mind, building documentation and interfaces and systems that reflect judgement instead of template gravity. The machine can get you to competent. It cannot tell you what deserves to exist.

As more clones saturate the internet and worldwide media, people will recognise the true point of difference, because it makes them feel something. If they ain’t feeling it, then your idea might not be as good as you think.

William Morris was right about the human cost. The response to mass production isn’t to reject machines. It’s to insist that the human hand still matters; that the maker’s presence in the work is not an inefficiency to be optimised away, but the very thing that gives the work its value.

Master your craft properly, or you’re just spreading the virus.

References

1. V&A, Introducing William Morris; Morris, Useful Work versus Useless Toil (1884); On Machinery, as reported in the press (1889). ↩

2. Benjamin, W. The Work of Art in the Age of Mechanical Reproduction, especially sections II–III. Harry Zohn translation, collected in Illuminations (Hannah Arendt, Ed.). ↩

3. Clayton, M., Royal Collection Trust. Leonardo’s study of anatomy. ↩ ↩²

4. Vatican Museums. Ceiling, Sistine Chapel. Work began in 1508 and was completed in 1512. ↩

5. Michelangelo, sonnet IV, “On the Painting of the Sistine Chapel,” to Giovanni da Pistoia. The Sonnets of Michael Angelo Buonarroti, translated by John Addington Symonds. The excerpt and closing phrase use the same translation. ↩

6. National Gallery, Leonardo in Verrocchio’s Workshop: Re-examining the Technical Evidence; Metropolitan Museum of Art, Michelangelo’s First Painting. ↩

7. Metropolitan Museum of Art. The Art of Renaissance Europe: A Resource for Educators, “The World of the Artist and the Patron.” ↩

8. Hintze, A., Proschinger Åström, F., & Schossau, J. Autonomous language-image generation loops converge to generic visual motifs. Patterns, 7(1), 101451. Published online December 2025. ↩

9. Shumailov, I. et al. (2024). AI models collapse when trained on recursively generated data. Nature, 631, 755–759. ↩

10. Kazdan, J. et al. Collapse or Thrive? Perils and Promises of Synthetic Data in a Self-Generating World. Preprint, revised March 2025; presented at NeurIPS 2024 workshops. ↩

11. Doshi, A. R. & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. ↩