AI Art and the Value of Execution

There is a fundamental disconnect in the debate over AI art, and I think much of it comes from the relationship different creative disciplines have with execution.

Art and engineering are cousins. We tend to treat them as entirely different pursuits, but both are fundamentally acts of creation. An engineer imagines a system that does not exist and builds it. An artist imagines an experience, image, sound, or story that does not exist and gives it form. Both involve taste, judgment, iteration, technical ability, and the translation of an abstract idea into something real.

The difference is where they tend to place value in that process.

For an engineer, execution is usually a means to an end. There can be tremendous craftsmanship in a beautiful implementation, and engineers absolutely develop preferences about how something should be built. But the purpose of engineering is ultimately to make the thing work. Nobody should be impressed that I manually wrote ten thousand lines of code if the same system could have been expressed more clearly in one thousand. Doing more work is not inherently virtuous. If a new tool allows me to build the same thing faster, or allows me to build something that I previously lacked the time or specialized knowledge to create, that is usually an improvement.

Art has a more complicated relationship with execution because execution can itself be part of the work.

This is why I think the claim that AI will destroy excellent art is fundamentally mistaken. Excellent art has never consistently followed the most efficient available method of production.

Practical effects are still used even when a scene could be created digitally. Film photography survives despite digital cameras being vastly more convenient. Musicians still record through old amplifiers, use analog synthesizers, or deliberately preserve imperfections that modern production software could remove in seconds. Traditional animation retains an appeal that cannot be reduced to whether newer techniques can produce more frames more cheaply. Something like Gunbuster is inseparable from the physical, hand-produced character of late-1980s cel animation; surviving production cels are literally artifacts of that process.

None of this is practical in the narrow sense. That is precisely the point.

Rick Rubin describes this idea well in The Creative Act: “Whether imposed by design or by necessity, it’s helpful to see limitations as opportunities.” Elsewhere, he writes that “the imperfections you’re tempted to fix might prove to be what make the work great.”

A limitation is not necessarily a defect in the creative process. Sometimes the limitation is what gives the work its character.

This is something technological arguments about art often miss. If art were simply a competition to produce the most technically perfect image as efficiently as possible, photography should have killed painting. Digital photography should have killed film. CGI should have killed practical effects. Perfect digital modeling should have eliminated all desire for handmade objects.

None of those things happened.

Instead, when a more efficient technique appears, the older technique often changes roles. What was once necessary becomes a choice. And once it becomes a choice, the fact that an artist deliberately chose the more difficult or constrained medium can itself become meaningful.

AI therefore does not make painting pointless any more than a synthesizer makes a piano pointless. They are different ways of creating, carrying different limitations and possibilities.

Where AI does pose an enormous threat is to a different category of art: art whose execution was never particularly important to the person purchasing it.

Stock photography is the clearest example. A company looking for a generic image of smiling coworkers around a conference table usually does not have a deep attachment to the photographer’s creative process. It needs an image of smiling coworkers around a conference table.

The same applies to enormous amounts of commercial illustration, background music, advertising, corporate video, generic copywriting, low-budget visual effects, and other creative work. There is also plenty of large-scale entertainment that is produced within such narrow commercial constraints that much of its creative labor begins to resemble industrial production. The fifteenth interchangeable promotional image for a franchise movie may technically be art, but very few people are buying a ticket because of the personal artistic expression of the person who created that image.

AI is extremely well positioned to consume this category of work.

But that does not mean artists are irrational for being frightened by it.

The uncomfortable problem is that commercially unimportant art is often what allows artists to make important art.

A musician may write advertising music to pay rent while working on an album nobody has agreed to finance. An illustrator may spend most of the week producing forgettable commercial graphics and use that income to paint on weekends. A filmmaker may work on disposable corporate productions while trying to get an independent film made.

The commercial work might have little artistic integrity of its own, but economically it matters enormously.

This is where I think much of the AI art argument becomes confused. There are really two different questions being asked.

The first is whether AI threatens art.

The second is whether AI threatens the economic system by which artists have traditionally supported themselves.

Those are not the same question.

I am skeptical of the first claim and far more sympathetic to the second.

The difference also explains why engineers have generally reacted to AI differently.

Software engineers certainly worry about employment, and plenty dislike AI-generated code. But there is much less cultural attachment to performing every part of the implementation manually. If I can describe a tedious integration and have an AI produce most of it correctly, I do not feel that the resulting system has become less authentic. I care whether it works, whether I understand it, whether its design is sound, and whether I can maintain it.

More importantly, AI dramatically increases the range of things I can create.

Engineers are often generalists surrounded by fields in which they have varying levels of competence. I might understand distributed systems extremely well but be mediocre at frontend development. I might know exactly how I want a tool to behave but have no interest in spending a week fighting CSS. I may understand an infrastructure problem but not know the exact syntax of a library I’ve never used.

Historically, each of those gaps placed a limit on what one person could reasonably build.

AI makes those gaps much smaller.

That does not eliminate expertise. I still benefit from knowing when the machine is wrong, what architecture makes sense, and what I am actually trying to accomplish. But I can increasingly spend my time deciding what should exist rather than manually executing every detail required to make it exist.

For an engineering generalist, this is incredibly empowering. My creativity is not diminished because I wrote fewer individual lines of code. My effective creative range has expanded because I can now build things that previously would have required more time, more specialization, or more people.

And this is where the argument comes back around to art, because the exact same effect can exist there.

The difference is simply scale.

A large artistic project has almost never represented the literal, unfiltered vision of one person. Consider a film. A director may have an extraordinary idea for what that film should be, but the final work passes through cinematographers, editors, composers, actors, production designers, visual-effects artists, sound designers, writers, and hundreds of other people.

Those people are not merely human tools executing the director’s instructions. They are creators themselves.

The composer may understand the emotional meaning of a scene differently from the director. The cinematographer may frame something in a way the director never imagined. An actor may fundamentally alter a character through a performance. An editor may discover that the movie works better when scenes are rearranged.

This is often wonderful. Collaboration can improve art precisely because other people do not perfectly share the original creator’s vision.

But collaboration is also partly a consequence of practical necessity.

A person who wants to make a film usually cannot compose an orchestral score, perform every role, operate every camera, construct every environment, create every effect, mix every sound, and edit the entire production at a professional level. Even if one person theoretically possesses all of those skills, there are not enough hours in a lifetime to execute every part of an ambitious project alone.

Money solves that problem by purchasing specialization.

AI changes that equation.

Imagine an independent filmmaker who knows almost exactly what music should accompany a scene but cannot compose and cannot afford a composer. Historically, the options were limited. Find a collaborator, compromise on the music, license something that is merely close enough, or abandon the idea.

An AI capable of translating that person’s musical intent into a usable score changes what the filmmaker is capable of expressing.

The result might be worse than hiring an excellent composer. In fact, it probably often will be. But that comparison misses the relevant alternative. The independent filmmaker was never choosing between AI and John Williams. The choice may have been between AI and no original score at all.

The same is true for animation, effects, concept art, sound design, editing, and eventually entire categories of production.

That has the potential to create something genuinely interesting: projects with a level of individual creative control that previously would have been economically impossible.

There is no guarantee that this produces better art.

If anything, the history of art gives us plenty of reason to be skeptical of the idea that eliminating creative resistance automatically improves a work. The Star Wars prequels are an amusing example. George Lucas gained vastly more technological and creative freedom than he had while making the original trilogy, and the results demonstrated that constraints, collaborators, and people willing to challenge a creator can sometimes improve the final product.

A collaborator is not merely an expensive obstacle between an artist and his vision.

Sometimes the other person’s different vision is exactly what the work needs.

AI therefore creates a strange tradeoff. It can give one person far greater control over a work while simultaneously removing some of the independent creative minds that would previously have shaped it.

Sometimes that will produce something sterile and self-indulgent.

Sometimes it may allow someone with an extraordinary idea and almost no money to create something that simply could not have existed before.

That possibility is difficult for me to see as inherently hostile to art.

In fact, it looks remarkably similar to what AI is already doing for engineering generalists: moving the boundary between imagination and execution.

This is why I think the most compelling criticism of AI art is ultimately economic rather than artistic.

AI does not eliminate the reasons people value human creativity. It does not make deliberate limitation meaningless. It does not make physical technique meaningless. It does not eliminate the appeal of seeing what another human being can accomplish with a particular medium. And it certainly does not guarantee that someone with unlimited generative capability will produce anything worth experiencing.

What it does threaten is the enormous layer of commercial creative labor in which the customer primarily wanted an output and the artist’s process was incidental.

Unfortunately for artists, that work has historically paid a lot of bills.

At the same time, the exact technology destroying some of those jobs may radically lower the cost of creating ambitious independent work.

That is the contradiction at the center of AI art.

It can weaken the economic infrastructure that has traditionally supported artists while simultaneously increasing the creative power available to an individual artist.

It can reduce the value of execution where execution was merely a cost while making previously impossible forms of individual expression economically feasible.

And it may force us to recognize something that engineers have been comfortable with for a long time: sometimes craftsmanship resides in the execution, and sometimes craftsmanship resides in knowing what ought to be built.

The interesting question is not whether AI can produce art.

It plainly can produce artifacts that occupy many of the roles we have historically assigned to art.

The more interesting question is what happens when execution ceases to be the primary constraint on creation.

For commercial creative work, the answer may be painful.

For genuinely individual creative work, we may only be beginning to find out.

Originally published on Substack.