Hollywood’s Downturn Is Turning Creative Know-How Into AI Training Data
The assignments provide paid work in a contracting production market, but turn experience in scheduling, development, and transcription into systems that may eventually perform pieces of that work.
Listen to this story
The audio brief
Story brief
3 key pointsA shrinking Hollywood labor market is supplying AI companies with highly specialized production judgment, not just basic labels. Through firms such as Mercor and Micro1, experienced creatives are evaluating scheduling, pitch-deck, budgeting, transcription, and coordination tasks for models tied to major developers including Anthropic and OpenAI. Pay ranges from $12 to $200 an hour, creating a short-term income...
- 01
Experienced creatives earn $12–$200 an hour training models on screenwriting, scheduling, budgeting, and production workflows.
- 02
Los Angeles shoot days fell 48% from 2021 to 2025, increasing the supply of experienced workers for AI-training gigs.
- 03
Micro1 sought producers with five-plus years’ credited experience, offering up to $85 an hour for evaluation tasks.
Hollywood writers, directors, and producers are taking AI-training assignments that ask them to teach models parts of their own craft, from planning shoots to preparing pitch materials. The work can offer immediate income, while raising a harder question: what happens when specialized production knowledge becomes training material for systems built to reproduce it?
How production expertise becomes model training
The assignments go beyond simple data labeling. Creatives give feedback on AI attempts at tasks for which they would ordinarily seek paid work, supplying the judgment needed to distinguish a plausible output from a usable one. Mercor, Micro1, and Handshake are among the AI-training companies connecting workers to this work; such agencies have contracts with major AI companies including Anthropic and OpenAI.
The tasks being broken into teachable parts
- Ruth Fowler, a screenwriter and producer, trained an AI to build a two-day shoot schedule from production materials. It had to account for permits, location hazards, daylight, personnel, and child-protection requirements.
- Fowler also trained a system to assemble pitch decks for film and television ideas, work associated with a production executive.
- A Los Angeles documentary director trained an AI on difficult video transcription, separating overlapping speech, music, and crowd noise while identifying speakers and accents.
A new buyer for skills amid a production slump
The market for this work is opening as entertainment employment contracts. Los Angeles shoot days fell 48% between 2021 and 2025, according to FilmLA Research. U.S. employment in motion-picture and sound recording declined from 450,000 in July 2022 to 326,000 in May 2026, according to Bureau of Labor Statistics figures cited by the Guardian.
Micro1 advertised for producers with more than five years of credited experience to author evaluation tasks around budget reconciliation, scheduling changes, and vendor or crew coordination, paying up to $85 an hour. For freelancers between commissions, that can be a practical bridge: one anonymous television writer said AI-training work had kept her afloat after commissions failed to rebound following the 2023 writers’ strike.
The employment outcome remains unsettled
Workers disagree over what their contributions will enable. The documentary director compared the job to being handed a shovel to dig the grave of his profession. Screenwriter Jody Wheeler, by contrast, said models could generate ideas but were not close to producing Oscar-winning scripts. Netflix said it used AI in 300 of its 1,000 titles in 2026, showing the technology is already present across a substantial share of its output. Whether these particular training exercises replace jobs, augment crews, or do both is still unresolved.
Sources
- theguardian.com‘Digging the grave of my profession’: the Hollywood creatives training AI to do their jobs
Loading discussion...
Reader comments
Newest comments first. Replies stay oldest first.