This is a piece I originally wrote as an Instagram story in November 2025.

If you asked me to name the hottest concept in AI right now, the answer would be, without a doubt, “agents.” At this stage, agents are mostly specialized for programming: they design their own workflows and solve problems on their own, functioning a bit like an outside contractor you hand a job to.

This concept will probably make its way into the art world before long. But can it actually replace art? And if it can’t, does that mean artists can rest easy?

Open Source and the Art World

I majored in design as an undergraduate and studied computer graphics in grad school. The most foreign thing I encountered while learning computer science was the spirit of “open source.” In this world, people pour enormous amounts of time into building things without getting paid for it, and then publish them on GitHub for anyone to freely use and modify.

The art world, by contrast?

I can’t generalize, but the art world is far stingier when it comes to sharing techniques. Even now, in an era where anyone can produce a stunning image with AI, you’ll routinely see Instagram posts promising to teach you their “prompt techniques”—just leave a comment and I’ll DM you.

Aura

Where does this difference come from? Several explanations are possible, but I think it comes from the “aura” of a work of art. An image that anyone can easily generate is, paradoxically, no longer “cool.” The novelty I felt when I first made a Ghibli-style image faded fast, and now nobody uses a Ghibli profile picture except for moms and dads. An increase in reproducibility is, in effect, the collapse of aura.

So even if AI agents advance to the point where you can just toss them a reference image and they operate Photoshop on their own to finish the picture, I believe the artistic value of that image will still depend on who made it and what they made it for.

The Problem of Data

There’s also a fundamental reason, from the standpoint of data, why AI agents will struggle to fully replace artistic production. For an agent to learn, it needs detailed workflow data tracing how a piece of work gets made. The reason coding agents could be deployed first is that programming work all happens inside the digital domain, is easy to turn into data, and the documents themselves are lightweight.

The workflow of art, on the other hand?

For one, just obtaining the data is a hurdle. Who would want to share their artistic workflow? That’s my secret sauce! Artistic masters who are already financially free, in particular, are almost never going to reveal their process—even if you offer to pay. You’d be lucky not to hear “So you think you can buy me?” But without that data, the workflows an agent can learn are bound to be superficial.

The format of the data is a problem too. Artistic work isn’t made of a single, uniform data type the way text is. It’s a structure where image, video, sound, material, space, context, personal experience, emotion, and abstract concept are all tangled together. Even if these can be physically quantified, it’s hard to standardize them in a way that preserves the meaning of an artistic judgment. When we can’t even agree on whether to unify a single category like 3D data under OpenUSD, what on earth would we standardize all the sensory and conceptual variables of art with?

So Can Artists Rest Easy?

So can artists rest easy?

No.

How many people within the art world are, every single day, making work that is genuinely so distinctive it couldn’t be captured in data, that proposes a new kind of sensation, or that overflows with “aura”?

The truth is that most artists are engaged in variations on existing forms, commercial visual work, or repetitive image production. This domain doesn’t require aura—and it’s precisely here that AI agents unleash their overwhelming destructive power.

Art that requires aura won’t be replaced. But the vast majority of visual work that doesn’t require aura—branding, illustration, content imagery, popular styles, decorative visuals—will be done by AI agents far faster, far more cheaply, and with a boring-level of consistency. In other words, the fact that AI can’t replace art doesn’t mean the art market is safe.

What today’s artists should fear is not “Can AI make art?” but “Can AI do most of what human artists have been doing, better?” And to put it coldly, the answer to the latter is already in.

AI may not, “for now,” replace the essence of art, but it will rapidly replace the productive role most artists have played—image labor. The change coming isn’t merely technical innovation; it’s a reorganization of the entire creative ecosystem. The tiny minority capable of producing aura will stand out even more, while the rest get swept into a form that’s indistinguishable from one another, lost in the enormous wave of images that agents churn out.

The future of art and artists will split, ever more extremely, between these two strata—the stratum of irreplaceable aura, and the stratum of fully replaceable production.

So What Now?

So what now? Fine, I get the situation—but what am I actually supposed to do about it?

Two directions come to mind right away.

The first is to go with the flow: actively embrace the technology and train yourself to become a more efficient producer of art. It’s probably the most realistic direction, but it also leaves a bitter aftertaste. Were “art” and “production” ever words that belonged together?

The second is to cultivate your own “aura.” Easy to say, but this is the comfortable, breezy kind of advice. If it were easy, everyone would do it. The process is extremely hard, and it’s questionable whether everyone can even reach the destination they’re aiming for.

I’m sorry that neither one looks like a clean answer. Still, neither direction seems entirely hopeless either. My guess is that most of us will end up walking somewhere in the middle of the two roads. I’m rooting for the future of everyone in the arts who walks that path.