Photography didn't kill painting, and AI won't kill writing.
In 1839, the painter Paul Delaroche reportedly looked at a daguerreotype and declared "painting is dead from this day."
He meant it.
A machine that could capture a face with complete fidelity had arrived, and he saw no way painting survived that.
He was wrong, but not stupidly wrong. He was diagnosing a real threat to a real part of painting: the commissioned portrait, the record-keeping, proof that a person existed. Photography absorbed all of that, but it ate the commodity layer whole.
What Delaroche missed was that painting was two jobs dressed as one.
Scroll any professional forum right now, and you will find writers certain AI has ended writing, marketers insisting AI-generated copy is slop, and HR departments quietly repricing the writing they used to pay for.
The fear is not stupid. Something real is happening. But the panic makes the same mistake Delaroche made.
Writing looks like one job, when it has always been two.
The first is high-frequency, high-volume, defined by coverage: product descriptions, summaries, SEO pages, email sequences, routine explainers, boilerplate.
Here, the limiting factor was never the writer's talent. It was time and cost.
The second is slow, specific, and irreplaceable by definition: the argument that changes someone's mind, the essay that explains a year of your thinking, the paragraph that lands because only one person on earth could have written it from what they observed.
Same role on the resume, different skill set, different market, entirely different threat model.
AI owns the first job. An MIT study of 453 professionals found AI cut routine writing time by 40% and improved quality scores by 18%.
Ted Chiang called ChatGPT "a blurry JPEG of the web" in 2023: lossy compression, useful for content mills, but not for the original thought.
A year later, he went further: a story takes something like ten thousand choices, and a prompt supplies maybe a hundred.
The model fills the rest by averaging what everyone else chose. He conceded the commodity tier to AI explicitly.
The mechanism: a NeurIPS 2025 paper tracking 70-plus language models found they collapse to the same handful of metaphors and sentence structures.
Intra-model similarity above 0.8 in 79% of cases. The researchers called it an "artificial hivemind."
RLHF enforces regression to the mean by design. The same process that keeps a model safe to deploy is what prevents it from having a point of view.
This maps directly to how anyone who has built automated systems thinks about the work: the decision worth a human's attention gets a human.
The task that can be characterized, repeated, and measured gets automated, and this is not because you distrust the tool, but because you understand what it is doing.
High-frequency, low-judgment tasks run on the machine. The calls where being wrong means something irreversible stay with a person.
Two jobs. Match the tool to the job.
The fear earns more than acknowledgment, tbh. Commodity writing was a real career path. Content agencies, SEO shops, junior copywriting roles, the content-mill tier of freelance: that work is being eaten, and it is not coming back.
The writers who built careers on volume are not overreacting. They are reading the market correctly.
So is every editor who got laid off when their publication decided AI could handle the summary posts.
The threat to that tier is real.
The question is whether conflating it with the other tier helps anyone make a better decision about where to put their energy.
So, in short, the panic conflates two different jobs that share a job title.
AI took the first. It is genuinely good at it, and pretending otherwise does not help.
But it structurally cannot do the second because the second requires the author to have been somewhere, to have seen something specific, and to have a position they can arrive at and defend.
Photography did not kill painting. It freed painting to become Impressionism.
AI won't kill writing. It'll free it to sound like an authentic voice again.
Wyt?