In 2022, two years after GPT3's release, Nothing, Forever was released. One of the more socially interesting and compelling early pieces of AI generated art, it was a procedurally generated legally distinct version of the sitcom Seinfeld, presented as an unending twitch stream.

When it was taken down in February of 2023 for perhaps not being legally distinct enough, it achieved viral notoriety. People enthusiastically piled in to marvel at its very existence and enjoy the bizarre nature of its rough edges. Like its low-poly art, the show's content was the product of still nascent AI technology, and the uncanny valley of its inhumanity was part of the attraction.

In mid 2025, a band whose photos felt suspiciously AI generated started to make waves on Spotify. The Velvet Sundown released 3 albums in about as many months, and had amassed a large following of over a million monthly listeners. Someone by the pseudonym of Andrew Frelon ran a fake Twitter account for the band and was solicited for interviews by dozens of journalists, passing off his opportunistic fraud as a "red-teaming" security exercise.

The band's actual music is anodyne 70s revival rock1. It's not spectacular, but unlike Nothing, Forever, it is not immediately obvious to a general listener that its twangy guitars are being strummed by six fingered men.

That was a year ago.

Here in 2026, @nicksaraev, entrepreneur and self-promoter who runs an AI business self help course has contributed to the "X is so over" discourse with this bit of AI psychosis:

[ ..] Spotify, Apple Music, and streaming in general is over.

AI music is now very, very good. You just don't know yet because all AI music circulating today is from ~2-3 generations ago. [..]

Suno v6 pro et al has solved infinite personalized music.

Infinite personalized music. We'll all get nothing, forever.

This phenomenon isn't limited to music or art. Software is my real area of expertise, and AI is a software product. It's from the software angle that I have really concluded that Saraev is right about one thing: AI output can be very good now.

I thought it was possible that AI would overcome its hallucinating slop era, but I didn't think it was inevitable2. Hallucinations are inherent to the design, but if their reliability reached a certain threshold, then their shenanigans would become tolerable for a lot of use cases. AI is perhaps better at generating code than at generating anything else, though perhaps part of this is a function of how little we care about code. Still, the idea that it's hit some threshold of utility is becoming conventional wisdom in the software world.

People who agree include nearly everyone I know personally in the industry, who it must also be said also hate what their jobs have become. It includes a broad gamut of technology thought leaders, from a cloud computing pioneer, an experienced infosec expert and entrepreneur, and a major figure in the burgeoning fascist prick software movement. They're all saying the same thing because it's impossible to ignore.

They also agree that software is not safe from becoming nothing, forever. Per tqbf:

My Mac menu bar and task switcher are cluttered with icons for programs I conjured for myself, with English as my programming language.

What I did on my computer, any power user can do on theirs. Will do on theirs.3

and:

[..] The core purpose of a modern operating system is to partition different applications off from each other, and carefully control how they can communicate.

That [..] makes less sense in the world we’re heading to, where most of the software we’re carving up fiefdoms for has the same provenance. And it makes almost no sense in a world where every application is malleable, subject to growing new limbs at any moment on the whims of its creator.

Currently, the good stuff AI generates is good because it is heavily curated by skilled human creators.

If you just ask AI to make a cartoon for you, chances are it will be so derivative of existing works that you'll get copyright strikes. An AI generated "trailer" for a non-existent film called Lord of the Force: A New Fellowship is as straightforward an example of slop as you'll see, because its creators had no creative vision, and their AI's beige amalgam of the history of cinema was a poor substitute for inspiration.

But, as Ben Affleck described in an interview to GQ, you can train AI on your own creative output, use it to produce small samples, or as an assistant in a creative process that you drive yourself. The difference between these projects and something like heavily sample-based music starts to blur.

When you decide to do something, you make all sorts of assumptions and micro-decisions while doing it. When you tell an AI to do something, it has to do this too. The more control you relinquish to it, the less reliably its output will match your internal vision.

This is why most of the best AI music is some kind of mashup, and why there is a booming cottage industry of classic game decompilation projects. If the AI has a very detailed vision to execute against, like a hardware specification and the original machine code, or samples comprising an artist's entire body of work, it can fill those gaps with something good. But ask the AI to write you a new Jazz standard, and you will get something utterly soulless.

How long that human spark is going to remain vital to getting good results out of an AI is an incredibly difficult question to answer, but I think the tsunami of AI generated personalized content being predicted is vastly over-estimated.

For those trying to drive that content stream themselves, I think they will find translating what they want into something actionable by AI to be more difficult than the pitch makes it sound, even as it improves and fills the gaps with better decisions. Sure, AI might be able to build a more detailed model of what makes a Truffaut film than you can, but it won't zero in on your vision of "Breakfast at Tiffany's but make it La Nuit Américaine". As a toy, a fascination, people won't mind. But fascinations are fleeting.

At its peak virality, Nothing, Forever had over 14,0001 concurrent viewers. Today, it averages 6.

Network effects are another huge impediment.

Weirdly, despite understanding virtually nothing else about anything, the NFT bros understood that part of the value of art is social.

I can appreciate Starry Night as an iconic masterpiece, but if it were not something that persisted in the world for others to also enjoy, it wouldn't become a part of me. It says something that the fractured, algorithm dominated world we live in today produces so few of the cross-cutting cultural touchstones of days gone past that Nothing, Forever had to riff on a nearly 30 year old sitcom.

Art can be almost entirely social in value. Viral internet moments can make people watch anything. They create bewildering levels of short term celebrity and/or infamy. You can blame online virality for hawk tuah meme coin rug pull being an intelligible English phrase.

Will bespoke content streams exist? Are people going to fire up the clanker to build the exact menu applet they want? Of course. Same as it ever was, really.4

A lot of music these days is simply consumed instead of listened to. The music that we love isn't like this, but in this headphone dominated world, is that the only music we listen to?

I have my headphones on right now, and there has been music coming out of them for some time, but I couldn't tell you what it is. In a new age of infinite AI driven musical content, Lo Fi Girl is a vulture at the feast. If you are a session musician that produces a lot of cool jazz records, I'd be incredibly concerned.

But on the whole, I think it's far more likely that after an initial explosion of people producing their own content, they will tire of it, even though AI has virtually eliminated most of the effort of doing so. The real effort is in having the idea. We're not content to sit with only our own ideas, and we're not content to keep them to ourselves.

On the software side, so much of what normal people use software for today requires the resources, authority, and trust of an organization. Yeah you can write your own clock, but are you going to slop up your own American Express app? Are you going to post to your own microblog?5

The printing press created more books than ever before, but this created many more readers than it did writers. The internet allowed anyone to publish anything to virtually everyone, but there are still more consumers of internet content than there are producers of it.

I just don't see AI flipping this imbalance.

  1. If you want good found footage from that era, you should go listen to Death instead.

  2. I do not believe in the inevitability of technological progress. Moore's Law, the idea that transistor densities will double every ~18-24 months, has been remarkably resilient, but this universe has laws, and Moore's law will eventually crash up against their shore. Every exponential in the real world is just an S curve where we haven't yet made it to the second knee. I tend to be overly pessimistic in how soon that bend will come. However, there is clearly still blood in whatever stones they're squeezing at the frontier AI labs. The idea that AI content quality has reached its peak is not one that should be taken seriously.

  3. As a long time Linux sufferer user, I was doing this the old fashioned way, before it was cool.

  4. You probably expected the Talking Heads, but I'm sorry. This is about the real-life Sharon Apple herself, Hatsune Miku, who/which is almost 20 years old.

  5. Are you going to write long and meandering essays on your own blogging platform?

Oct 13