"You spend years writing a book. Then within hours (minutes?) of publishing it, AI copies the idea, rewrites and improves parts of it, strips its origin and distributes it worldwide for almost zero cost."
That's shadcn describing the machine he built his career atop. Not the printing press copying the book — the machine that copies the creative work, improves parts of it, and redistributes it before you've finished your launch post.
He ends not with a diagnosis but a question: "What happens then to the will to create?"
His replies rushed to supply an answer. It arrived almost instantly, and it was almost unanimous: taste can't be copied.
"Execution just lost its scarcity. Taste didn't." "Copying takes only the artifact, it cannot absorb judgment, context, and the ability to decide what's worth making in the first place." "The founders I've seen survive commoditization all had one thing in common: they had a specific point of view, not just a product."
It's a comforting consensus. It's also being falsified in real time by the tools these same people open every morning.
Because here is what taste looks like once a builder decides to work with a coding agent: a file named DESIGN.md. The "design source of truth." The document, per the practitioners writing them, that "tells Claude what the product should look and feel like."
Read that back slowly. The thing everyone just agreed was uncopyable — taste, point of view, the feel of a product — is now being deliberately written down in machine-readable prose so a model can act on it.
That's the trap folded inside the reassurance. The value that "can't be copied" turns out to be the exact value you're now compelled to serialize.
And DESIGN.md is only the tidiest example. shadcn asked a sharper version of the same question himself: what happens when your roadmap, your changelog, and your launch announcements become input prompts?
They already have. A roadmap is a document of judgment — this before that, this instead of that, this not at all. A changelog is a record of which decisions survived contact with reality. Both are public. Both are now scrapeable context.
So the defense collapses on its own terms. Taste can't be copied, everyone agrees — while producing an ever-expanding corpus whose entire purpose is to make taste legible, transferable, executable by something that isn't you.
The copying machine that thinks doesn't need to reverse-engineer your judgment from your output anymore. You're handing it the judgment directly, formatted, versioned, and committed to main.
This is the part the "free" argument gets right and then walks past. shadcn's real point was never that copying is cheap. It's that copying used to cost the copier something — you had to understand a thing well enough to reproduce it, and that understanding was itself a form of skill.
AI removes that cost. And a DESIGN.md removes it further, because now the copier doesn't even have to infer the understanding. It's been pre-written.
So if the artifact can be copied, and the documented taste behind the artifact can be copied, what's actually left?
Here the more useful reframe is one nobody in shadcn's replies quite reached: the artifact was never the proof in the first place.
"The introduction of AI means we can no longer rely on any artifact as proof of thinking," as one writer put it. "Authorship is no longer the thing. Thinking-ship is."
For your entire working life, the artifact stood in for the thinking. If you shipped the code, launched the product, wrote the essay, we assumed the cognition happened somewhere behind it. The output was a receipt.
That assumption is dead. A polished artifact now appears in seconds. The receipt can be printed without the purchase.
Which means protecting originality was never about protecting the output — the output never proved anything on its own. It was always about protecting the act of judgment that produced it. We just couldn't see the difference until copying got cheap enough to separate the two.
Protecting originality was never about protecting the output. It was always about protecting the act of judgment that produced it.
And this is exactly why the DESIGN.md paradox is real but not fatal. A DESIGN.md is a snapshot. It captures the judgment you had reached by the moment you wrote it.
What it cannot capture is the next judgment — the one that decides, tomorrow, that half of yesterday's DESIGN.md was wrong.
The document is a frozen output of an unfrozen practice. Copy the file and you copy the position. You do not copy the positioning — the continuous act of deciding, revising, and knowing which thing is worth making next.
That distinction is the whole game. The same writer who declared authorship dead described the test: he asked an AI to role-play his client and interrogate him, and he couldn't answer. "I had generated the deck without thinking."
The artifact existed. The thinking hadn't. And no one could tell from the deck — including, for a moment, its own author.
That's the future of originality, and it's brutally simple. The artifact will always be plausible. The only proof left is whether you can still do the thinking when someone removes the artifact and asks the next question.
A DESIGN.md can't pass that test on your behalf. It can only tell an interrogator what you already decided — not defend why, not revise it under pressure, not choose differently when the market moves.
So why does anyone still create when the copy is free and even the reasoning behind the copy is downloadable?
Not because output is scarce — it isn't, and pretending otherwise is the losing position. The reassurance crowd is quietly betting that scarcity migrates from execution to taste and then holds still. It doesn't hold still. Taste written down is taste commoditized.
The will to create survives somewhere else entirely: in being the person who knows which thing is worth making next. That position is not an asset. It's a practice.
An asset can be documented, versioned, and stolen. A practice has to be re-earned every time the context changes, which is constantly.
This is the quiet concession inside shadcn's framing. When you treat authorship as a shippable artifact — a book, a component library, a DESIGN.md — you've already handed the machine everything it needs, because a finished artifact is by definition copyable.
When you treat authorship as an unfinished practice, there's nothing to hand over. The judgment lives in the next decision, which hasn't been made yet, which no file contains.
Every prior copying machine killed a monopoly on distribution, not creativity. This one is different only in that it also strip-mines the documented reasoning behind the work. But it still can't reach the one thing that was never in the document.
So write the DESIGN.md. Publish the roadmap. Ship the changelog. Feed the machine all of it — the artifacts were never where you lived.
Just don't confuse the file with the faculty that wrote it. The moment you can't answer the next question without reading back your own notes, you've become copyable — not by the machine, but by anyone, including it.
The will to create was never about owning the copy. It was about being the one who knew what to make. That job description hasn't changed. It has only stopped accepting receipts as proof.