Your AI campaign looks like a campaign. That is the problem.

Somewhere in an agency right now, a creative team is typing a prompt into a video model, and the prompt says "cinematic, 35mm, shallow depth of field, golden hour". They are asking a machine that has never held a camera to pretend it held a camera.

Jack Conte has a name for that, and it is the one everybody uses and nobody defines. He is the CEO of Patreon and one half of Pomplamoose, and in his SXSW 2026 talk he spent 58 minutes on a question most AI keynotes skip. What is slop?

"Slop is that lazy phase after the introduction of a new medium where we're all just trying to reproduce existing things that are honestly just better if the people do them."

The important word is phase. Not a property of the tool, not a verdict on the people using it. A phase. And he has 130 years of examples to show that every new medium went through the same one.

Film spent seven years pretending to be theatre

The Lumiere brothers built the cinematograph in 1895 and nobody knew what it was for. So the first filmmakers used the only reference they had. They built stage sets, parked the camera where the front row would sit, and had actors gesture at the back of a room that was not there. Maxim Gorky watched one in 1896 and described a world dipped in monotonous grey.

Then in 1902 Georges Melies made A Trip to the Moon. Fourteen minutes, cuts, changing locations, a rocket in the moon's eye. It worked because it stopped trying to be a play. It used the one thing film could do that theatre could not.

Recorded music ran the same loop. The first producers put a band in a room and tried to capture a gig. Thirty years later the Beatles were looping two-bar drum breaks and running vocals through organ speakers on Revolver, and George Martin called the studio "a new canvas on which they had to paint". The medium started when someone stopped imitating.

Now read the prompt again. "Looks hand-drawn." "Looks like it was shot on film." We are pointing the wide lens at the stage.

The unions were not stupid

This is where the talk stops being reassuring, and the part to keep for the next time someone argues about AI and jobs.

James Petrillo, president of the American Federation of Musicians, claimed recorded music was ruining the jobs of 60 per cent of his members. So he banned it. The Down Beat front page on 1 August 1942 read RECORDING STOPS TODAY, and the ban held for years.

He was not wrong about the jobs. When the Vitaphone brought sound to cinemas in 1927, cinema owners did one sum. The machine cost $9,000 to $15,000. A 15-piece pit orchestra cost about $50,000 a year. Owners fired the band and the machine paid for itself in months. By 1930 the union counted 22,000 musicians out of work, while weekly cinema attendance went from 50 million to 90 million. Both numbers are true. Both belong in the same sentence.

Charlie Chaplin hated talkies and held out for thirteen years, until The Great Dictator in 1940. Conte does not mock him for it. He wonders whether Chaplin felt caught between being left behind and changing his work in a way he did not believe in. If you have ever been a copywriter staring at a prompt box, you know the feeling.

The synthesizer refuses to resolve

The Hammond Novachord showed up at the 1939 World's Fair with 163 vacuum tubes, roughly 500 pounds, and an immediate union ban. The fear was justified. In 1941 every Billboard number one was recorded by an orchestra. Today, essentially none. A 2008 Stanford study found 46 of the 63 largest US orchestras running deficits.

The same instrument gave us Stevie Wonder, Prince, Radiohead and Daft Punk. Conte brings a synth player on stage mid-talk and asks the audience to hold both facts at once without picking a side. Do not decide whether the synthesizer is good or bad. Hold all of it.

Underneath it, he says, is Baumol's cost disease. You cannot squeeze productivity out of a string quartet; four people playing Schubert takes exactly as long as it did in 1828. The synthesizer did not kill the orchestra so much as answer an economic problem the orchestra could never solve. The next time someone says AI "killed" a discipline, ask what economic problem it answered.

Happily ever after is not a real place

The most useful section for anyone running a business is the least quoted one. Conte walks through his own career as a graph. YouTube plus iTunes downloads made Pomplamoose a full-time act. Then a YouTube redesign stopped the growth. Then streaming replaced downloads. Then Patreon took his evenings, 8am to 10pm at the office and filming from 11pm to 3am, for two years, until the band went dormant for three. He read that as proof he had failed as an artist.

Then weekly uploads rebuilt it to over a million monthly Spotify listeners. Then COVID. Then vertical video, which took them from roughly 6 million monthly views to 30 million. He puts both bars on a slide and labels the comparison "Totally Meaningless", because short-form views do not convert to Spotify and carry no ad money.

A career is not a mountain with a summit. It is an infinite sine wave. Change does not mean death. He admits it is the tritest thing anyone has said on a SXSW stage, and he is probably right, and it is still the thing he wishes someone had told him at 27.

If it is fair use, why are you paying Disney?

Conte's case against the AI companies rests on what they do, not on what they claim. They claim training on creators' work is fair use, and at the same time sign multimillion-dollar deals with Disney, Conde Nast, Vox and Warner Music. If it is legal to just take it, why pay anyone? And why pay them and not the illustrators, musicians and writers?

He is careful about what he is not saying. He runs a tech company. Patreon would be dead in three years if it did not adopt new technology fast. He does not think the world owes artists consistency. The ask is narrow. Pay people when you use their work. "If we're going to task people with figuring out how to build data centers in space, then we ought to task people with figuring out how to pay creators for their work."

Three things a model cannot do yet

The last fifteen minutes are the part to argue with. They are three separate arguments, and two of them are stronger than the one everybody repeats.

• Leaps. Train a music model only on recordings up to 1959. Does it get to the Beatles without the Beatles? Then to Prince? Then to Nirvana without Cobain? Culture moves in jumps, and the current models interpolate.

• Risk. Comedians work outside the socially acceptable range, and that is why we watch. Great artists put themselves on the line. "If an LLM says some crazy shit, it's not interesting to me. What's it risking? Who cares?"

• Scarcity. Karen X Cheng put it best. Art is not what is easy to make, it is what is hard. Invisible selfie stick videos got millions of views until everyone found the setting on their Insta360. Now they get nothing.

Then the concession that makes the rest credible. Conte hates the word "content" because it puts a Kendrick Lamar record and a video about unclogging a toilet in the same category. His split is media that carries information and media that carries connection with another conscious being. He concedes the first kind to AI outright. Dinner jazz in restaurants, instructions for hanging a picture, headline summaries. He argues only about the second.

What this means for the prompt box

Most AI work in advertising right now is the wide lens pointed at the stage. A generated spot that looks like a spot. A generated illustration that looks like an illustration. It is fine, it is fast, and it is worse than the thing it copies, because the thing it copies is better at being itself.

"Can the model make our ad" is the wrong question. The right one is what can this medium do that a camera and a Cintiq cannot. A thousand personalised endings. A campaign that rewrites itself against the weather. Something nobody has a name for yet, which is the point. Nobody has made the Trip to the Moon of this yet. That is not a reason to despair. It is the only open brief in the building.

Watch the talk. Then go and find the thing it is not a play of.