Artist Jeff Koons summed up our current moment decades ago, with a sculpting series known as Balloon Dog.
Over ten feet tall, cast in mirror-polished stainless steel, perfectly, absolutely, aggressively smooth. Not a single imperfection on its surface. No texture. No grain. No evidence of a human hand having touched it. No seam where the mold closed. No nick where a tool slipped. The things look as though they materialized from nothing, which is precisely the point. An orange Balloon Dog, one of five Koons made, sold for $58.4 million in 2013 and became, for a time, the most expensive work by a living artist ever auctioned.
Of course it did. Balloon Dog is the perfect expression of the dominant aesthetic of our time. The aesthetic of the smooth.
Han's most beautiful argument is about smoothness. It is the one that keeps me awake. The one I cannot quite dismiss. Smoothness. Not the metaphorical smoothness of efficient processes, but the literal, aesthetic, cultural smoothness that has become the signature of this century.
Look around. The iPhone: a slab of glass so featureless it could have been grown, not made. The Tesla dashboard: a single screen, no buttons, no knobs, no tactile resistance of any kind. One-click purchasing. Frictionless checkout. Seamless onboarding.
The word "seamless" is used as a compliment. It shouldn't be – or at least, it shouldn't be used that way without understanding the cost of seamlessness. A seam is where two pieces meet. Where the joints show. Where the maker's hand is visible, the construction legible, the artifact revealing its own process.
A seamless garment hides its construction. A seamless experience hides its complexity. And when you hide the construction, you hide something essential about the thing: the labor that made it. The decisions that shaped it.
The friction that tested it. Han traces this aesthetic through contemporary culture with the eye of someone who can see what others have stopped seeing.
Botox: the smoothing of wrinkles, which are the records of expression, of having lived in a face long enough to mark it. Instagram filters: the elimination of blemish, shadow, asymmetry, and everything else that makes a face specific, located, human.
In each case, the friction is removed. In each case, something real goes with it.
Before AI, writing software was a sequence of productive failures. You conceived a function. You wrote it. It did not work. You received an error message, specific and unhelpful and sometimes maddening, that told you something had gone wrong without telling you what. You read the error. Examined the code. Hypothesized. Tested. Failed again. Read documentation, often badly written documentation that assumed you already knew the thing you were trying to learn. Asked on Stack Overflow. Someone answered dismissively. You tried again. Eventually, hours or days later, the function worked.
In those hours or days, something had happened that was not visible in the final code. You had come to understand the function. Not intellectually, but in your body. The kind of understanding that lives in the fingertips, in the pattern-recognition apparatus that builds itself only through repeated exposure to failure and correction. You understood the function because you had struggled with it. The struggle was the understanding. The friction was the learning.
Claude removes this friction. You describe the function. Claude writes it. It works. You move on.
The code is correct. It may even be better than what you would have written. But you have not understood it in your body.
Think of it instead as a geological process. Every hour you spend debugging deposits a thin layer of understanding. The layers accumulate over months and years into something solid, something you can stand on. When a senior engineer looks at a codebase and feels that something is wrong before she can articulate what, she is standing on thousands of those layers, each one laid down through friction, through the specific resistance of a system that did not do what she expected.
Claude skips the deposition. The surface looks the same. The knowledge has been transferred, not earned. The friction that would have built the understanding has been smoothed away, and with it, the specific depth that can only be built through struggle.
This pattern extends beyond coding, because Han's argument is not about code. It is about the nature of experience in a world that has decided friction is always a cost and never a benefit.
Consider the lawyer who uses AI to draft briefs. The briefs are competent. They cite the right cases, make the right arguments, organize the analysis in a structure the judge expects. But the lawyer who produced them has not read those cases – at least not in the way you read a case when you are searching for the argument that will save your client, reading not for information but for the one sentence that might shift the entire framework, the dissent that contains a logic the majority didn't consider. That kind of reading is slow and uncomfortable and full of dead ends, and it is also the process by which a lawyer becomes someone whose judgment you would trust with your life.
The AI drafted the brief. The lawyer reviewed the output. The client may be well served today. But the lawyer has not deepened their understanding of the law. They have extracted a result without undergoing the experience that would have made them better at their work next year.
Consider the student who uses AI to write an essay. The essay is articulate, demonstrates understanding of the material, and might even receive a high grade. But the student has not thought the thoughts that the essay represents. The friction of wrestling with an idea until it yields its meaning, the kind of thinking that only happens when you are stuck, when the words will not come, when the argument resists your attempts to make it cohere? That friction has been bypassed. There is ample evidence that desirable barriers support a healthy learning process.
The essay exists. The understanding does not.
Consider me, writing this book with Claude. The connections between ideas, the structural clarity, the range of reference are enhanced by the collaboration.
But I am certain the enhancement has come at a cost.
It could be that the ease of producing text with Claude has allowed me to avoid the specific, painful, productive kind of thinking that happens only when I am alone with a blank page and nothing works. It could be that the smooth surface of AI-assisted writing conceals a loss of depth that I cannot yet see because I am inside it.
The most dangerous aspect of smoothness is that it is self-concealing. The smooth does not announce itself as a loss. It announces itself as a gain. The code that works without struggle feels like progress. The brief that writes itself feels like efficiency. The essay that arrives without pain feels like intelligence.
And because the loss is invisible, it compounds. Each frictionless interaction reinforces the expectation of frictionlessness. Each time you accept AI output without questioning it, the questioning muscle weakens slightly. The developer who has used AI for six months finds the idea of debugging manually not just tedious but intolerable, as though she has been asked to walk somewhere after learning to fly.
The tolerance for friction atrophies, and with it, the capacity for the thinking that only friction produces.
I caught this myself, multiple times, as I polished this book. The Deleuze failure I described in Chapter 7, where Claude produced a passage that sounded like insight but broke under examination? That’s the aesthetics of the smooth in action. The prose was polished, but the reference was wrong, and the smoothness concealed the seam where the argument fractured.
The aesthetics of the smooth is a cultural system in which the absence of resistance has become the standard of quality, and in which the things that resistance produces – depth, understanding, embodied knowledge, the satisfaction of having earned something difficult – are quietly disappearing. What haunts me is not the loss itself, but the speed at which we stop noticing it.
The algorithmic feed delivers content with such precision that you never encounter anything that disturbs you. The AI assistant drafts your response before you have decided what you think. The recommendation engine learns your taste and serves you more of it, which means you encounter less of what you did not already prefer, which means the capacity for surprise decays right alongside the capacity for friction.
The smooth world is a comfortable world. It is also a world in which the muscles you need most are the muscles you use least, and you never sit with discomfort long enough to learn from it.
I am going to let this chapter end there. In the discomfort, in the loss, in the weight of a diagnosis I cannot entirely refute.
Because the next chapter will bring data. Real evidence, from real researchers, about what actually happens when AI enters the workplace. And the data will confirm some of what Han fears and complicate much of what he assumes.
The counter-argument will only be convincing if you have felt the weight of the diagnosis first. So, sit with it. Let it breathe.
Understand what is wrong, and we’ll try to get to what is true.