About ten years ago I was wandering through the Musée Rodin in Paris. I am not an artist. I went in with the naive assumption most of us carry into a gallery: Rodin made the Rodins.
Then I discovered the studio behind the name.
Rodin was deeply involved in the conception and modelling of his sculpture, but the physical realization of major works often depended on a network of assistants and specialist craftspeople. The Musée Rodin explains that he employed skilled practitioners to enlarge and reduce models, marble carvers to translate plaster into stone, foundries to cast bronze, and assistants such as Camille Claudel who worked on difficult details. The hand that struck every blow of the chisel was not always Rodin’s. Yet we still say: “a Rodin.”
That discovery stayed with me because it disturbed a simple idea of authorship I had never examined.
Then, a few days ago, I read a New York Times profile titled “The Incredible Art of Being Jeff Koons”. The same question returned, but in a form that feels remarkably relevant to the age of AI.
At one point Koons had more than 100 people working in his studio, including teams hand-copying old masters. Today the operation is smaller, but the system remains exacting: digital models, scans, standardized lighting, paint-by-numbers codes, external foundries and specialist fabricators. Koons told the Times that his systems are intended to ensure that every gesture, colour and shape is the way he wants it.
Who, then, made the art?
The easy answer is: his team. The more interesting answer is: Koons authored it.
That distinction may help us understand one of the most uncomfortable questions facing writers now: if an AI helps research, structure, edit and even produce sentences, who is the author?
The false choice: purity or abdication
The argument around AI and writing has hardened into two camps.
At one extreme is abstinence. A “real” writer should write every word. AI may be allowed to fix spelling or perhaps find a source, but the moment it generates prose the work becomes suspect.
At the other extreme is abdication. Give the machine a topic, ask it to research, outline, draft and polish 1,500 words, glance over the result, then publish it under your name.
The first confuses authorship with keystrokes. The second confuses supervision with authorship.
The scale of the temptation is easy to understand. In a randomized experiment published in Science, Shakked Noy and Whitney Zhang gave 453 college-educated professionals realistic writing tasks. Access to ChatGPT reduced average completion time by 40 percent and increased assessed quality by 18 percent. That is not a marginal productivity improvement. It is a reason every writer, marketer, consultant and knowledge worker is being forced to rethink how work gets made.
But speed is not the same as authorship.
From maker to architect
Humans have been moving up the abstraction stack for centuries.
We moved from muscle to machines; from doing every task to operating the machine; from operating the machine to designing the system. The builder becomes the architect. The craftsperson becomes the creative director. The founder who once needed a department can increasingly orchestrate a network of software and AI agents.
AI accelerates that move because it can automate not only physical labour but parts of cognitive labour.
For writers, the ladder might look like this: sentence maker → editor → director → architect of meaning.
There is enormous leverage in that progression. One person can explore more research, test more counterarguments, surface more connections and move from idea to published work faster than ever before.
But there is a trap in the metaphor of “moving up.” Higher leverage does not automatically mean higher mastery.
Rodin could delegate because Rodin knew sculpture. He had modelled clay, studied bodies, understood proportion and developed taste through practice. That accumulated craft gave him the capacity to look at work produced by someone else and say: no, that is wrong.
The same is true of a great editor, architect, chef or conductor. Direction is strongest when it rests on an internal model built by doing the work.
AI now allows us to leap up the leverage ladder before we have climbed the mastery ladder. That creates a new figure: the fragile director. They can generate impressive-looking work at extraordinary speed but may not possess the craft required to tell whether it is actually good.
Making is also thinking
This is why I resist the idea that the ideal future is one in which humans stop doing and spend all their time directing machines.
Part of what I love about art is technical excellence: seeing what a human hand has learned to do with marble, pigment, a brush, wood, sound or light. Sometimes the execution is not merely a delivery mechanism for the idea. The execution is part of the idea.
The same applies to writing.
A sentence can contain information, but it can also contain rhythm, restraint, surprise, personality and years of practice. We admire Orwell or Didion not merely because of what they thought, but because of the precision with which they learned to express thought.
The danger is that AI can give us the appearance of mastery without the apprenticeship that produces judgment.
A 2026 review in Trends in Cognitive Sciences describes this broader problem as cognitive offloading. Offloading work to AI can impede skill acquisition or contribute to skill decay, although the authors stress that the outcome depends on how the technology is used. That qualifier matters. The issue is not AI itself. It is what we choose to stop practising.
Automate friction that wastes life. Preserve friction that develops you.
Transcribing a two-hour interview by hand is mostly friction. Searching 100 documents for a quote can be friction. Reformatting citations is friction.
But wrestling with an argument can be formative friction. Finding the sentence that says exactly what you mean can be formative friction. Writing an introduction before asking AI for one can expose what you actually think.
The AI creativity paradox
The research increasingly suggests that AI creates a trade-off rather than a simple win or loss.
In a Science Advances experiment, Anil Doshi and Oliver Hauser gave writers access to generative-AI story ideas. Writers who could request up to five AI ideas produced stories rated 8.1 percent more novel and 9 percent more useful than the human-only group. The biggest gains went to writers who started with lower creativity scores.
But the stories also became more alike. AI-assisted stories moved closer to the average story in their condition. With access to one AI idea, the increase in similarity represented 10.7 percent of the similarity-score range found in the human-only group.
That paradox has since become harder to dismiss. A 2025 study of 2,200 college admissions essays found that human-written essays added new semantic diversity roughly two to eight times faster than base GPT-4 essays as the number of essays increased. And a 2026 meta-analysis spanning 19 studies and 61 effect sizes found a small but statistically significant homogenisation effect in human-AI co-creation.
AI can make each of us better while making all of us more similar.
That is where “AI slop” enters the story.
In 2025, Merriam-Webster named “slop” its word of the year and defined it as low-quality digital content produced, usually in quantity, with AI. The phrase is useful, but I think slop is a symptom rather than the disease.
The deeper problem is abdication.
AI slop appears when we outsource not only production but curiosity, experience, point of view, taste and judgment. The machine supplies the topic, the structure, the examples, the language and sometimes even the conclusion. The human becomes a publishing endpoint.
The result can be grammatically clean and intellectually empty.
A new definition of authorship
This is why Rodin and Koons matter to the AI writing debate.
They show that authorship has never required the author to perform every physical act of production. Art has a long history of workshops, apprentices, assistants, foundries and specialist fabricators. What matters is the nature of the contribution and the degree of creative control.
Even copyright law is moving toward this distinction. In its 2025 report on AI and copyrightability, the U.S. Copyright Office concluded that using AI as an assistive tool does not prevent copyright protection. Human-created selection, arrangement or modification can qualify. But simply providing prompts is not, by itself, enough to establish authorship of AI output.
That is a legal standard, not a complete philosophy of writing. But the direction is useful.
Authorship is not “I touched every word.”
It is closer to five responsibilities: Origin — why does this work exist? Intent — what am I trying to say? Direction — what should be researched, included, excluded or challenged? Judgment — is this true, interesting, beautiful, useful and mine? Responsibility — am I prepared to put my name behind it and defend it?
None of those requires typing every sentence. All of them require being present.
The two ladders of AI-assisted creation
I now think creators need to climb two ladders at once.
- The first is the leverage ladder: maker → operator → director → architect.
- The second is the mastery ladder: novice → apprentice → craftsperson → master.
AI can rocket us up the first ladder. It cannot automatically carry us up the second.
The dangerous position is high leverage and low mastery: the fragile director.
The exciting position is high leverage and high mastery: the master-director. That is Rodin with a studio. It is the architect who understands construction. It is the editor who has written thousands of pages. And it may be the strongest model for the AI-age writer.
The boundary I am trying to draw
I am still working this out in my own writing.
Sometimes I have abdicated too much. I have supplied a topic or headline and let AI run too far. Other times I have written the opening, supplied the lived experience, directed the research, challenged the argument, rejected language, moved sections and edited heavily. Increasingly, I think of the second approach as authorship rather than purity.
The percentage of AI-generated words is a poor test.
A better test is whether the work would exist in substantially the same form without the human behind it.
Did the piece begin with something I noticed, experienced or genuinely wanted to understand? Did I decide what question mattered? Did I challenge the evidence? Did I choose what belonged and what did not? Could I explain and defend the argument without opening the AI chat? Would another person giving the model the same headline have produced essentially the same article?
And one more question may matter even more: Am I still practising the craft that allows me to judge the machine?
I don’t want AI to free me from writing. I want it to free me from unnecessary labour so I can spend more time on observation, thought, story, craft and judgment.
The future should not be a civilisation of people who have forgotten how to make things but have become excellent at requesting them.
Nor should we romanticise unnecessary labour merely because humans once had to perform it.
The better destination is the master-builder: hands capable of making, a mind capable of designing, judgment capable of directing, and technology capable of multiplying all three.
Use AI to expand thought and amplify expression. Do not let it decide what you mean.
Or even more simply:
Delegation can expand authorship. Abdication abandons it.
The deeper question: what will you do with all this leverage?
AI can help us write faster.
It can help us research more deeply, explore more options and produce at a scale that was impossible a few years ago.
But that creates a new problem.
The more capability we gain, the more important it becomes to know:
What do I actually want to create?
What is worth my attention?
What deserves my time, energy and commitment?
The danger is not only that AI starts writing for us.
It is that we become surrounded by so many possibilities that we lose sight of our own direction.
That is part of why I’m building Zyrro.
Zyrro is designed to help you understand the patterns behind who you are, what energizes you, what matters to you and which paths may be worth exploring next.
Not to hand you a fixed answer.
Not to tell you what your purpose is.
But to help you make better choices in a world where AI can generate almost infinite options.
Because the real opportunity of AI is not simply to produce more.
It is to give us more leverage to become more intentional about what we choose to make, pursue and become.
AI can amplify your capabilities. Zyrro is being built to help you decide where to point them.
If that sounds useful, join the Zyrro waitlist and follow the journey as we launch.
Research & source links
- Musée Rodin — Multiples, fragments, assemblages
- Musée Rodin — Marble carving
- Musée Rodin — Camille Claudel
- The New York Times — The Incredible Art of Being Jeff Koons
- Noy & Zhang (2023), Science — productivity effects of generative AI
- Doshi & Hauser (2024), Science Advances — creativity and collective diversity
- Moon, Green & Kushlev (2025) — homogenizing effect of LLMs on creative diversity
- de Rooij & Biskjaer (2026) — meta-analysis of homogenisation in human-AI co-creation
- Cash et al. (2026), Trends in Cognitive Sciences — cognitive offloading and skill decay
- U.S. Copyright Office — Copyright and Artificial Intelligence, Part 2
- AP — Merriam-Webster’s 2025 word of the year: slop
The post AI Didn’t Kill Authorship. It Changed What Authorship Means. appeared first on jeffbullas.com.
* This article was originally published here
