AI is making creators more valuable, not less

Written by Priyanshu Mishra, CEO of Lumienzo. Sources are named inline. Last reviewed 15 September 2026.

In June 1998, in the Spanish city of Leon, Garry Kasparov sat down to play Veselin Topalov, and this time both men had a computer at their elbow. Kasparov used a program called Fritz 5. Topalov used one called ChessBase 7.0. Kasparov had invented the format himself, a year after Deep Blue beat him, and the idea behind it was quietly brave: stop asking who wins, the man or the machine, and ask what the two of them can build together.

Here is the part worth keeping. One month earlier, Kasparov had played the same man at ordinary fast chess and won four games to nothing. Not a close match. A rout. In Leon, with a machine on each side of the table, the six main games finished level, three all.

Kasparov described it years afterwards in The New York Review of Books, on 11 February 2010: "A month earlier I had defeated the Bulgarian in a match of 'regular' rapid chess 4-0. Our advanced chess match ended in a 3-3 draw. My advantage in calculating tactics had been nullified by the machine."

Sit with that for a second. The computer did not make Kasparov stronger. It cancelled out the one thing he was better at than anyone alive, and the moment it did, the man he had just beaten 4-0 could hold him to a draw. The contest did not end. It moved. It moved to preparation, to judgement, to nerve, to knowing which of the machine's suggestions was worth anything.

One honest footnote, because people argue about this match. The Week in Chess reported in 1998 that Kasparov went on to win a sudden death playoff after those six games, so he took the overall event by a point. The six main games were level, and that is the part this page is about.

Something close to that is happening to creative work now, and it is not the story in the headlines. We build AI for creative work, so read this knowing that. The losses come before the good news here, every figure names who published it and when, and the last section lists the popular numbers we refused to use because they fall apart when you check them.

The rule underneath that chess match

David Autor, an economist at MIT, wrote the general version of it in the Journal of Economic Perspectives in 2015: "when automation or computerization makes some steps in a work process more reliable, cheaper, or faster, this increases the value of the remaining human links in the production chain."

He named the idea after the O-ring, the small rubber seal that failed and destroyed the Challenger shuttle in 1986. Think of a job as a chain of links where any single one can break the whole thing. Make nine of those links cheap, fast and reliable, and the tenth link does not become less important. It becomes the only thing that decides whether the chain holds. The person holding it gets more valuable, not less.

Now look at a brand deal as a chain. Writing the plan. Building a shortlist. Agreeing a price. Drafting the contract. Chasing the invoice. Reading the comments afterwards. A machine can do all of those quickly, and it is getting quicker. The one link it cannot do is be the person whose audience believes them. That is where the value goes.

What is genuinely being lost, before any of the good news

You are allowed to be frightened. Real people have had real work taken from them, and a page that skips over that has not earned the right to reassure you about anything else.

In January 2024 the Society of Authors in the United Kingdom asked its members what had happened to them, and 787 people replied. Among the illustrators, 26 percent said they had already lost work to generative AI. Among the translators it was 36 percent. Those are not forecasts about the 2030s. Those are people describing the last year of their working lives.

It gets more uncomfortable. Xiang Hui, Oren Reshef and Luofeng Zhou tracked a large freelance marketplace and published what they found in Organization Science in November 2024: after ChatGPT arrived, writing freelancers got about 2 percent fewer jobs and about 5.2 percent less money every month. And here is the detail that should stop anyone reciting a comfortable line about quality winning out. The best rated freelancers were hit hardest, not protected.

Ozge Demirci, Jonas Hannane and Xinrong Zhu reported in Management Science in 2025 that job posts for writing and coding fell about 21 percent in the eight months after ChatGPT, and that posts for image work fell about 17 percent once the image generators landed. Erik Brynjolfsson, Bharat Chandar and Ruyu Chen at the Stanford Digital Economy Lab reported in August 2026 that employment for people aged 22 to 25 in the most exposed jobs is running about 19 percent below where it would be if it had tracked the less exposed jobs. The youngest people are carrying the most of this, which is the least fair way it could have landed.

And some trades really do end. The US Bureau of Labor Statistics reported in September 2025 that one hour photofinishing, the shop that used to develop your holiday pictures, went from 17,320 jobs in 2000 to 437 in 2024. That is a fall of 97.5 percent. James Feigenbaum and Daniel Gross studied what happened when the telephone company replaced human switchboard operators across more than half the American network between 1920 and 1940, and published it in the Quarterly Journal of Economics in August 2024. Younger women moved into other work. Older women largely did not. Even when the total number of jobs recovers, the individual people do not always recover with it.

Now notice what every one of those has in common. They are all about work sold as output: a file handed over, a task completed, a print developed, a call connected. Not one of them is about a person with an audience.

Now read the same studies to the end

Almost nobody quotes the second half of these papers, and the second half is where the hope lives.

Demirci, Hannane and Zhu, in that same Management Science paper in 2025, found that the jobs which survived in the automation-prone categories were "of greater complexity and offer higher pay". Fewer jobs. Harder jobs. Better money.

The Stanford team found the line that actually divides the winners from the losers, and it is not your job title. It is how the tool gets used in your line of work. In their words: "In occupations where AI is used more to complement workers, employment is flat or rising, particularly among more experienced workers."

And for artists in particular, Christos Makridis of Gallup looked at earnings on 3 May 2026 and found that artistic jobs with higher exposure to generative AI are earning broadly what the less exposed ones earn. His summary is the sentence to carry around with you: "AI is changing how artists work, not whether they work."

The bank teller story, with the half nobody tells you

Cash machines did exactly what people feared they would. James Bessen wrote the numbers up in Finance and Development, published by the International Monetary Fund in March 2015: between 1988 and 2004, the number of tellers needed to run a branch in the average American city fell from 20 to 13.

Then came the surprise. A branch was now cheap to run, so the banks opened more of them, 43 percent more in urban areas, and the total number of teller jobs did not fall. The counting moved to the machine, and the person moved to the counter as someone who could open an account, fix a problem, arrange a mortgage, remember your name.

Bessen is careful about why that worked, and so are we. It worked because there was more demand for banking waiting to be served, and because tellers could learn the new work. Remove either of those conditions and the same automation gives you the opposite ending.

Autor wrote the sentence that decides which teller survived it: "a bank teller who can tally currency but cannot provide 'relationship banking' is unlikely to fare well at a modern bank." The story does not stop in 2004 either. The US Bureau of Labor Statistics now projects teller jobs to fall 13 percent by 2035. What finally took the job was the phone in everyone's pocket, not the cash machine.

For you the test has the same shape, and it is uncomfortably simple. If what you sell is a file, you are on the counting side. If what you sell is the reason a particular group of people watch you and believe you, you are on the relationship side, and that side has been winning for forty years of automation.

The most confident prediction in this field was wrong

In 2016 Geoffrey Hinton, who later won a Nobel Prize for his work on neural networks, said: "People should stop training radiologists now." He said it again in The New Yorker, published on 3 April 2017, and he reached for a cartoon to make the point. A radiologist, he said, was like the coyote who runs off the edge of a cliff and keeps running: "You're already over the edge of the cliff, but you haven't yet looked down."

He was the most qualified person alive to make that call, and he was wrong. Deena Mousa reported in Works in Progress in September 2025 that radiology was the second highest paid medical speciality in the United States, at an average of 520,000 dollars in 2025, and that American training programmes offered a record 1,208 places that year, 4 percent more than in 2024.

Why? Because reading the image was the rule shaped part, and machines did get good at it. Deciding what to do next, telling a frightened person in a quiet room, and carrying the responsibility for being wrong were not rule shaped, and they turned out to be most of the job. Hinton counted the task. He did not count the job. Hold on to that difference the next time somebody counts your task.

Two men from New Hampshire, and what they actually had

In June 2005 a chess website called Playchess.com ran a tournament with one rule: anything goes. Bring a grandmaster. Bring a supercomputer. Bring both. Teams of strong grandmasters entered with several machines each. Two Hydra supercomputers entered, each running on dedicated chess hardware.

It was won by two men who had never held a title in their lives. Steven Cramton was 28. He coached soccer at a prep school, ran the school snowboarding programme in winter, and coached a small chess team in spring. Zackary Stephen was 24, had a masters degree in statistics, and worked as a database administrator. They called their team ZackS. ChessBase reported on 19 June 2005 that they won the final 2.5 to 1.5, and that their ratings were around 1700 and around 1400, which in chess terms is a thousand points below the people they were beating. They played on three ordinary desktop computers.

People refused to believe it. A rumour went round that Kasparov was secretly feeding them moves, which Cramton answered in ChessBase in 2005: "Just to set the record straight, we did not have the help of any GM's, IM's or any titled players or any other players for that matter."

What they had instead was a way of working. They knew their programs the way a mechanic knows an engine. They knew which one to trust in which kind of position. They picked two or three candidate moves out of their own experience, split the analysis between them, compared what each program said, and argued it out. Kasparov, writing in The New York Review of Books on 11 February 2010, drew the conclusion himself: "Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process."

Read that plainly. The best player did not win. The best hardware did not win. The best method won. A method is something you can learn this month, which makes it the most encouraging fact on this page. Two details to keep the story honest: ChessBase reported on 11 June 2005 that both Hydra supercomputers were knocked out before the quarter finals, so ZackS never beat one face to face, and the team they beat in the final had a grandmaster on it.

And the honest ending to the chess story

The advantage those teams had was real, and then it faded. As the engines got stronger, the human contribution in a chess game shrank towards nothing, and the one job the human used to do, choosing which program to trust, is now done by another program. Tyler Cowen wrote on Marginal Revolution in February 2024 that in this kind of chess "the entity making those choices is now a program, not a human being". No tournament result settles it formally, so treat it as the accepted view among people who follow the game rather than as a measurement.

We are telling you that because the version of this story that leaves it out is advertising, and because the real lesson survives it. Chess is a closed world. One goal, fixed rules, a perfect scoreboard. That is exactly the kind of world machines end up owning.

Your work is not that world. Nobody can write down the objectively correct video to post on Tuesday. No scoreboard settles whether a joke landed, whether a brand deserves your audience's attention, or whether the thing you are about to make is worth making at all. Where the goal itself is a matter of judgement, the human link keeps its value. That is not a loophole. It is most of human work.

What to do with all this

Stop selling the file. Start selling the judgement. Anyone can generate a draft now, so a draft is not a product. What you decide to make, whether it worked, and what to do differently next time: that is a product.

Put the tools on the parts nobody sees. First drafts, admin, scheduling, the boring half of every deal. Shakked Noy and Whitney Zhang tested exactly that with 453 professionals doing real tasks from their own jobs, and published it in Science in 2023. Time fell 40 percent and quality rose 18 percent, but the sentence that matters is about shape rather than speed: the tool "restructures tasks towards idea-generation and editing and away from rough-drafting".

If you are early in your career, start today. Erik Brynjolfsson, Danielle Li and Lindsey Raymond followed 5,172 customer support workers through an AI rollout and published it in the Quarterly Journal of Economics in 2025. Productivity rose 14 percent on average, but new and low skilled workers improved 34 percent while the experienced ones barely moved. Anil Doshi and Oliver Hauser found the same pattern in creative work, in Science Advances in July 2024: among 300 people writing short stories, the writers who had scored lowest on creativity beforehand gained the most. These tools raise the floor. If you are standing near the floor, that is your news.

Then protect the thing that is yours. The same Doshi and Hauser study found that stories written with an AI idea were 10.7 percent more similar to each other than stories written without one. Individual work got better and the pool got narrower. Leonard Boussioux, Jacqueline Lane, Miaomiao Zhang, Vladimir Jacimovic and Karim Lakhani found the matching result in Organization Science in 2024, when 300 evaluators judged 234 business ideas: the ideas produced with a human guiding the model were more workable, and the human crowd was still the more original one. Originality is the thing you sell. Argue with the first draft. Never just accept it.

Where Lumienzo sits in this

We automate the middle of a deal on purpose. Lumi writes the plan and the first ideas, finds creators on verified audience data, agrees the price, handles the contract and the escrow through a regulated partner, pays on approval, and reads every comment afterwards. It does not make your video and it cannot make your audience. Every one of those links getting cheaper is, by Autor's finding, exactly what makes the person at the end of the chain worth more. See how it works for creators, or get started. Then go and make the thing only you could have made.

The figures we refused to publish

This subject is full of numbers that do not survive a check. If you see these quoted at you, now you know.

• "90 percent of online content will be AI-generated by 2026." Nobody measured this. It began as a personal forecast by the author Nina Schick in her 2020 book about deepfakes, a Europol report cited her in 2022, and the internet turned a citation of one person's guess into an apparent finding by a police agency.
• "Goldman Sachs says AI will destroy 300 million jobs." The March 2023 report says exposed, not destroyed. Exposed means some of the tasks overlap with what a tool can do. The same report put full substitution at about 7 percent of US employment and complementarity at about 63 percent.
• "AI destroyed 204,000 entertainment jobs." CVL Economics published that in January 2024, after asking 300 executives in late 2023 what they expected to happen by 2026. A survey of expectations is not a count of losses.
• "Two amateurs with laptops beat a supercomputer." The most repeated version of the story above, and it is wrong twice. ChessBase reported in 2005 that ZackS played on three ordinary desktop computers, and reported on 11 June 2005 that both Hydra supercomputers were already out of the tournament before the quarter finals.
• "Teams of humans and computers still beat the best engines at chess." They almost certainly do not, and we found no tournament result or study that settles it either way, so the page says what is believed rather than pretending to a measurement.
• "From today, painting is dead", said by Paul Delaroche in 1839. Almost certainly never said. The historian Stephen Bann traced the earliest trace of it to an account written in 1873, 34 years afterwards. Delaroche's documented view of the new photography was positive, and he went on painting until he died in 1856.
• "Hayao Miyazaki called AI art an insult to life itself." He said those words in 2016, in a documentary, about a procedural animation of a figure dragging itself along by its head, and his objection was about disability and pain. Usable image generators did not exist yet. He has said nothing about them.
• Creator AI-adoption percentages with no source, of the "86 percent of creators use AI" kind. Nearly all of them come from surveys run by the companies selling the tools, on panels those companies picked themselves.

Common questions

Will AI replace creators?

No, and the direction of travel is the opposite one. A machine is cheap at the steps whose correct answer can be written down before the work starts, such as a first draft or a shortlist. It cannot make a person that other people already trust. As the first kind of work gets cheaper, the second kind gets rarer, and rare things cost more.

Why would automation make a creator worth more?

Think of a job as a chain of links, where any weak link breaks the whole thing. The economist David Autor wrote it out in the Journal of Economic Perspectives in 2015: "when automation or computerization makes some steps in a work process more reliable, cheaper, or faster, this increases the value of the remaining human links in the production chain." Make nine links cheap and fast, and the tenth link is now the only thing that decides whether the chain holds. You are the tenth link.

But people are losing work to AI right now. Is that not true?

It is true, and we are not going to pretend otherwise. The Society of Authors in the United Kingdom surveyed its members in January 2024 and heard back from 787 of them: 26 percent of illustrators and 36 percent of translators had already lost work. Xiang Hui, Oren Reshef and Luofeng Zhou reported in Organization Science in November 2024 that writing freelancers on a large marketplace got about 2 percent fewer jobs and about 5.2 percent less money each month after ChatGPT arrived, and that the best rated freelancers were hit hardest rather than protected. What is being lost is work sold as output. What is holding is work sold as judgement and as audience.

Which creative people are actually exposed?

The ones whose product is a file rather than a following. David Autor put the test in one sentence about banking, in the Journal of Economic Perspectives in 2015: "a bank teller who can tally currency but cannot provide relationship banking is unlikely to fare well at a modern bank." Counting notes was the part a machine could take. Knowing the customer was not. If people follow you because of who you are, you are on the second side of that line.

What should a creator actually do about it?

Stop charging for the part a machine now does cheaply, and charge for the part it cannot reach. Put the tools on first drafts and admin, because Shakked Noy and Whitney Zhang found in Science in 2023 that generative AI "restructures tasks towards idea-generation and editing and away from rough-drafting". Then spend the hours you just got back on the audience relationship and on your own taste, because nobody can write down in advance what a good idea looks like.

If a brand can generate a video, why would it pay a creator?

Because the video was never the thing it was buying. It is buying the attention of people who trust a particular person, and that trust does not move to a face a computer drew. Christos Makridis of Gallup wrote on 3 May 2026 that earnings for artistic jobs with higher exposure to generative AI look broadly similar to earnings in jobs with lower exposure, and summed it up like this: "AI is changing how artists work, not whether they work."