America has shed more than 200,000 jobs across media, film, and the arts in four years, roughly 50,000 of them in the past twelve months alone. Joseph Politano’s Apricitas analysis, published September 27, puts the movie and sound recording sector at the center of the wreckage: more than 100,000 jobs gone, nearly one-third of the sector’s total, with summer employment hitting its lowest level since the 2008 recession and nearing a 30-year low. Written publishing lost more than 70,000 over the same stretch.
The obvious question is how much of this is AI. Politano’s honest answer is that nobody can prove it, and his piece is more valuable for admitting that than for the numbers, striking as they are.
The case is circumstantial, and Politano knows it
The evidence for AI as a driver is structural rather than direct. Every subsector of digital arts is contracting while in-person entertainment grows at a healthy pace. That split is suggestive. If the problem were a broad consumer pullback, live venues would be hurting too. They are not. What separates the losers from the winners is whether the work can be generated, copied, or distributed as bits.
The attribution problem is real. Media firms are, in Politano’s phrase, “decidedly coy” about their AI use, both to dodge public backlash and, more consequentially, to protect the legal basis of their copyrights. Admitting that a model drafted your script is a bad look and a worse litigation posture. Add Hollywood consolidation, offshored production, and the migration of audiences to social creators, and the causal picture gets muddy fast.
But the muddiness cuts one way. The sectors bleeding hardest are exactly the ones where generative tools are cheapest to deploy and hardest to detect. Politano cites the figure that 32% of workers in arts, entertainment, and recreation use generative AI to some extent, against a 62% average across all industries. That lower number is not reassurance. It means the displacement is early, not absent.
The Napster precedent is the part to read twice
Politano’s sharpest move is historical. He walks through Metallica versus Napster, the lawsuit the band won and the war the industry lost. Piracy destroyed the economics of recorded music, Spotify and Apple Music consolidated the catalogs, and the price of a song collapsed toward zero. Musicians survived by converting recordings into loss-leading advertisements for live shows. Tours lengthened, ticket prices climbed, and the job became closer to influencer than artiste.
Hollywood has no equivalent escape hatch. Politano is blunt: there is “basically no equivalent to live concerts that Hollywood can be used to ease the pain.” Movie ticket sales ended last year down 25% from their 2019 peak, cable revenue down 18% from its peak, and streaming has not closed the gap. That is before the flood of AI-generated video competing for the same attention.
The grimmer comparison is newspapers. US newsroom payrolls once employed roughly half a million people, more than the oil industry, and are now down nearly 85%. Magazines shed 66% of staff since the 1990s, book publishers 40%. The mechanism was unbundling: box scores to sports sites, stock tables to finance sites, forecasts to weather apps, ad revenue to Google. AI is the same unbundling applied to the production of the content itself, not just its distribution.
What the data cannot yet tell us
Here is where we part with the more alarmist readings. Politano’s own counterexample undercuts a clean narrative. The novel was dying, with fewer books sold in 2019 than at the turn of the millennium. Then TikTok arrived, BookTok turned reading into a social performance, and book sales surged 70% over six years. A short-form video app rescued the printed word. Nobody forecast that.
The lesson is not that AI will save creative work. It is that the second-order effects of a new medium are genuinely unpredictable, and confident forecasts in either direction are mostly vibes. What we can say is narrower and firmer: the current losses are concentrated in digital production, they are accelerating, and the tools that plausibly contribute to them are getting cheaper every quarter.
Why AI builders should care about this number
The creative-sector collapse is the first large-scale, real-world test of a claim the AI industry has made loosely for years: that generative models will augment workers rather than replace them. The macro data is starting to lean the other way, at least for the parts of the workforce whose output is a file rather than a performance.
That has consequences beyond labor economics. If AI-driven displacement in media becomes the accepted story, it strengthens the hand of the copyright plaintiffs, the licensing regimes, and the regulators now circling training data and model outputs. Politano notes that firms hide their AI use partly to preserve copyright claims. A visible body count in Hollywood and publishing makes hiding harder and litigation more likely.
For builders, the practical read is that the cheapest, most defensible AI products may be the ones that make human creative labor more productive in ways a client can document, not the ones that quietly replace it. The studios that survive the next four years will be the ones that can point to what the model did and what the person did.
The outstanding question is measurement. Nobody has a clean dataset separating AI substitution from consolidation, offshoring, and audience drift. Until someone builds one, the 200,000 figure will be cited by both sides of the argument, and the honest position is that we know the direction of travel without knowing the speed.
Politano’s piece ends mid-sentence in the version we read, cut off at “At the end of the day, t.” Fitting, in a way. The story is still being written, and the people who will be asked to prove what happened are the ones currently deciding whether to admit they used the tools.