The Numbers Behind the Creative-Class Downturn: 200,000 Jobs and the AI Question

Joseph Politano’s analysis at Apricitas Economics, first published September 27 and all over Hacker News this weekend, assembles the employment data behind a story usually told in anecdotes: the United States has lost more than 200,000 jobs in creative industries over four years, roughly 50,000 in the past year alone, a stretch the author compares in intensity only to the 2001 and 2008 recessions, except this time there is no general recession. The losses are concentrated with unusual precision: movie and sound recording, Hollywood proper, is down over 100,000 jobs, nearly a third of the sector and its lowest summer employment since 2008, and written publishing is down more than 70,000. Broadcasting, streaming, and graphic design fill out the losses. In-person entertainment keeps growing.

Why the in-person split is the strongest evidence

Politano is careful about attribution, and the care is what makes the analysis useful. Proving AI caused specific job losses is hard: media firms hide their AI use to avoid backlash and copyright exposure, and the losses are entangled with Hollywood consolidation, content offshoring, and displacement by social-media creators. But the digital-versus-physical split is the fingerprint. Every subsector whose output can be produced on a screen is shrinking simultaneously, while live entertainment grows, and the industries that lost to earlier technologies show the pattern’s limits: newspapers’ payrolls are down about 85% since the 90s and music never recovered its CD-era revenue after Napster, but musicians survived by pivoting to touring and merchandise, a fallback Hollywood does not have, with ticket sales down 25% from 2019 and cable revenue down 18%. A cited survey puts generative AI use at about 32% of arts and entertainment workers, below the cross-industry average, with a commenter noting a work-only series shows 42%, so adoption and displacement are not moving in lockstep yet.

What this means for technical readers

For the audience deciding where to build, three takeaways survive the caveats. First, the displacement arrived through workflows, not products: AI seeped into storyboarding, localization, drafting, and post-production invisibly, and firms cannot fully prevent even their own employees from using it, which is why the losses do not map onto any single AI product launch. Second, the labs read this data too: Politano notes Anthropic skipped image, video, and music generation, and that OpenAI abandoned Sora, suggesting the major labs judge pure art generation economically marginal, which implies the creative job losses to date came from cheap tools, not frontier models. Third, the macro impact remains modest, arts are a small sector and unemployment sits below the 2010s average, but the same analysis flags tech as the other clearest AI-affected sector, experiencing its longest labor-market downturn after roughly 300,000 annual gains in 2022. We have documented the same pattern in adjacent niches, from AI wiping out Kenya’s essay-writing industry to expertise not becoming optional; the creative-class data adds the aggregate that anecdotes could not.

What to watch

Three things will sharpen the picture. First, whether the ~50,000 annual pace continues or the losses decelerate as the most automatable workflows saturate; the data is now updating fast enough to watch in near-real time. Second, whether Anthropic’s and OpenAI’s apparent retreat from media generation holds, because frontier-model entry would change the slope of every curve in the analysis. Third, the spillover channel Politano flags as unpredictable: BookTok drove a roughly 70% book-sales surge over six years that nobody forecast, and the creative economy has repeatedly produced compensating categories from nowhere. The honest summary is the one the data supports: a historic contraction concentrated exactly where software touches creative output, growing at 50,000 jobs a year, with attribution improving and the largest models still sitting it out.