# Hollywood Was Sinking Before AI Arrived — Notes on Odd Lots Live in LA

Source: Realpha Blog (blog.getrealpha.com)
Original article and charts: https://blog.getrealpha.com/en/blog/oddlots-2026-09-25-hollywood-was-cooked-before-ai-and-now-it-s-only-g/

> Bloomberg Odd Lots, 2026-09-25, live in Los Angeles: writer Hayes Davenport tells the last decade through shoot days — from 40,000 to under 20,000. Educational notes, not investment advice, no stock recommendations.

Published: 2026-09-25
Locale: en
Tags: Hollywood, AI, content industry, industry structure, Los Angeles

TL;DR: Shoot days in Los Angeles have halved in a decade, and the line turned down before AI showed up; the thing squeezing content is free, not cheap.


![An empty Los Angeles soundstage at dusk, loading door half open, one shaft of light cutting through dust into the depth of the stage, palm trees outside](/covers/oddlots-2026-09-25-hollywood-was-cooked-before-ai-and-now-it-s-only-g-cover.png)

> By the Vermilion Bird Bridge, wild grasses flower;
> at the mouth of Blackrobe Lane the sun slants low.
> The swallows that nested in the halls of Wang and Xie
> now fly into the homes of ordinary people.
>
> —— Liu Yuxi, "Blackrobe Lane" (Tang dynasty, c. 826; translation mine)

On 25 September 2026, Bloomberg's *Odd Lots* recorded its first live show in Los Angeles at the Vermont theater, where hosts Tracy Alloway and Joe Wiesenthal interviewed writer Hayes Davenport (*Eastbound and Down*, *Family Guy*). One number carries the episode: shoot days in Los Angeles peaked around 40,000 in 2016 and have fallen to under 20,000 this year. That line started down in 2016, the pandemic took a cut out of it, the 2023 strikes took another — and AI still hasn't entered the writers' room. His line "we've been competing for years against the sum of all human creativity, produced for free — you can't scare us with AI" applies to fiction writing in TV and film. Below-the-line crews sit outside that shelter, and he says so himself: they're on a precipice.

## What the episode is about

The subject is what's left of Hollywood. Hayes is now creative director of LA Material and co-hosts *Hollywood Handbook*, and he spent 2021 to 2024 working in Los Angeles City Hall — so he can see both the inside of a writers' room and the street outside the municipal window. The hosts open with "is Hollywood screwed," and instead of answering yes or no he reaches for the industry's own unit of measure — shoot days — and lays out ten years. From there the conversation moves through production leaving town, rooms shrinking, where AI actually is today, and what happened to restaurants and nightlife once the production volume fell.

## The points worth keeping

**The ten-year line turned down before AI.** Shoot days in LA ran near 40,000 in 2016, easing to about 37,000 by 2019. After the pandemic collapse, backlog pushed 2021 up sharply, and 2022 came back to roughly 37,000. The 2023 strikes knocked it to 25,000; 2024 brought no recovery and landed at 23,000. The phrase going around town was "survive till 25," and instead this year came in under 20,000 — half of 2016. For that gentle 2016-to-2019 slide, Hayes points to streaming building entirely new networks while cable shrank underneath. His first jobs were Comedy Central shows, and Comedy Central doesn't really do new shows anymore.

![Line chart: LA shoot days slide from their 2016 peak down to the right, with a pandemic trough and a strike cliff along the way; the turning-point marker sits at the far left, while the point where AI gets blamed sits far to the right.](/figures/la-shoot-days-decade-line-en.svg)

**Work follows the tax break, and LA's stages finished building at the wrong time.** *Eastbound and Down* shot in North Carolina until the governor ended the break there; the same team took their next show, *Vice Principals*, across the border into South Carolina, and *Righteous Gemstones* stayed there too. Plenty of his friends shoot in Vancouver now. The one that surprised him most: Fox game shows and studio comedies are filmed in Dublin, Ireland — they fly Rob Lowe and an American contestant cast over, for tax and labor reasons. Meanwhile that 2016 peak financed a wave of stage construction, and the holes dug around 2019 are now finished buildings. Demand isn't where it was, so the studio real estate side has its own problem.

**Rooms are shrinking, and so is the one method a machine can't copy.** *Family Guy* had thirty writers when he was there. A mini room used to mean seven writers breaking off to go write one joke; seven is now a big room for a whole show. On the last show he helped with, the two creators had budget for one person to come in and outline the season for a couple of weeks, then they wrote it themselves. Post-strike terms require a room, once triggered, to hit a certain size for a certain number of weeks — which changed the math on opening one at all. What was the room for? There's a slot in the script for a joke and people pitch into it, and you hear what gets a laugh. On *Family Guy* you also had to do the voices: "if you can't do Lois, your joke will not get in."

![Three clusters of dots shrink from thirty to seven to one, each labelled with the size of writers' room it stands for.](/figures/writers-room-shrink-en.svg)

**A new show competes against the entire library, not the other network.** *Family Guy* memes are the lingua franca of the internet, which means every new show is up against it — something that wasn't true before. Hayes adds a second layer: the old shows compete with each other too, so *Family Guy* and *The Office* still command a real price while big chunks of the library keep sliding down in value.

![Bars on a shelf: the two tallest on the left are older shows that hold their value, a long run in the middle gets shorter and fainter as the library loses value, and the single bar on the right is a new show squeezing onto the same shelf.](/figures/library-shelf-never-clears-en.svg)

**Horror is alive, comedy is retreating.** A mid-budget horror film is commercially legible — a studio knows it will make money, a road Blumhouse opened with *Paranormal Activity*, made for almost nothing and returning millions. Comedy has no comparable upside or perceived theatrical pull, which is why Zach Cregger and Jordan Peele left team comedy. Hayes ties that to AI: comedy is the hardest thing for a model to do, and models can already write outlines and handle other genres. The category that resists replacement is the category shrinking in volume.

![Quadrant chart with difficulty of AI replacement on the horizontal axis and upside potential on the vertical: horror grows on the cheap-to-model, easy-to-replace side, comedy shrinks on the hard-to-replace but hard-to-model side, and an arrow shows creators moving from comedy to horror.](/figures/horror-comedy-quadrant-en.svg)

**The bar AI has to clear is free, not cheap.** His words: you talk about AI being cheaper, it is not as cheap as free. Fiction has spent years competing with the sum of everything humans post for nothing, distributed for nothing, sorted algorithmically to your taste — and we each have only so many consumption hours. Joe adds the same shape from his own trade: newspapers took it on the chin from Google News and everything after it for twenty years, and AI joins the back of that queue. One detail I enjoyed: every time a new model ships, Joe asks it for ten tweets in the style of Joe Wiesenthal. Usually wrong, but one came back with "the 10-year yield doesn't care about your feelings, and frankly it doesn't care about mine either." He wouldn't post it. He kind of liked it.

![A horizontal axis running from expensive to free: big-budget productions sit at the left, AI generation sits toward the right but still short of the end, and the far right is free output from everyone, with the uncrossed gap marked in between.](/figures/cheap-is-not-free-en.svg)

**The blast radius is much wider than the industry.** Production-side crews — hair and makeup, set design, grips, what the city calls below the line — sit right at the edge when it comes to AI, and for local employment that's the large problem. From City Hall, Hayes watched small businesses: restaurants taken to the edge by the shutdown, then shoved over by the production decline. A writers' room ordering lunch is somebody else's revenue. A busy lot means everyone walking around it eats nearby. Tourism moves too — *Insecure* brought parts of the city to life and people moved there to live that. The line in town is that Hollywood isn't the LA economy, maybe 20% of it. It also isn't a silo.

## Going further

### When a headline says AI is disrupting an industry, how do you check it

What I kept turning over was whether the timing lines up. The story version is: AI arrived, Hollywood fell over. The number version is that the line turned down in 2016, streaming replaced cable, then the pandemic and the strikes each left a hole, and AI still hasn't written a show. Those two versions point at opposite actions. If the cause is structural migration plus a tax race to the bottom, then waiting for the AI hype to deflate is waiting for nothing.

My habit now is to draw one measurable line before listening to any story, and to use the unit the industry itself uses — here, shoot days, not box office or subscriber counts, because shoot days are what move lunch orders, hair and makeup, and stage rent. Then one question: did the suspect in the story appear before the turn or after it? Anything that shows up after is an accelerant. I learned this the slow way, by reaching for a recent headline to explain a revenue drop and then finding the drop had started two years earlier.

![Two panels each show a declining line and a suspect marker: in the left panel the suspect appears after the turning point, so it can only be an accelerant, while in the right panel the suspect appears before the turning point, so it could be the main cause.](/figures/turning-point-vs-suspect-en.svg)

### When your product competes with your own back catalogue, the new rival isn't the thing to watch

The library point stayed with me. A new show has to beat *Family Guy*, which means supply never leaves — old work doesn't get pulled, doesn't depreciate to zero, and sits on the same shelf. Looking at a content company's library, instinct offers either "more content, so libraries are worth more" or "AI can generate, so libraries are worth nothing." Hayes gives a third shape that looks more like an actual business: the head titles hold price, the middle erodes piece by piece.

This isn't only a library problem. Your old model is still circulating used, your old course material is being passed around, your article from three years ago still ranks above today's — all of it competing for the same attention you're trying to sell into. To judge whether a company's back catalogue is an asset or a rival, I look for two numbers: whether the old items still convert (licensing, renewals, resale price), and whether acquisition cost on the new items keeps climbing. First one falling while the second rises means the catalogue has flipped sides.

![Two lines cross: revenue from old titles slopes down while the cost of acquiring viewers for new titles climbs, and the crossing point marks where the back catalog turns from an asset into a competitor.](/figures/stock-flips-two-lines-en.svg)

### When an industry halves, who bleeds first

The order Hayes saw: restaurants go first, because their buffer is thinnest. The month writers stop getting paid, the lunch orders stop. And AI's first wave doesn't land on writers either — it lands on crew. That gives me a reading order: shocks travel along the supply chain, and the first to bleed is the layer with the least pricing power and the thinnest cash cushion, which is usually a layer nobody is tracking.

![Three horizontal bars of differing thickness are stacked from thinnest to thickest, with an impact arrow striking in from the left and the bars labeled one, two and three in order, so the thinnest bar is punched through first.](/figures/who-bleeds-first-en.svg)

Which makes "Hollywood is 20% of the LA economy" a misleading comfort — it measures the industry, not the reach. Going back over my own holdings, I noticed every question I ask is about the company itself, and almost never about which of its suppliers or customers runs out of runway first. The Emily Sundberg detail makes it concrete: she's in LA, doesn't know of anything happening that night, stays in the hotel. The city has what he calls a DoorDash nightlife. Between "shoot days fell" and "nothing to go to tonight," not one step gets reported.

## Worth a look

- Bloomberg *Odd Lots*, 25 September 2026, live in Los Angeles (daily newsletter at Bloomberg.com/oddlots)
- FilmLA's annual shoot-day statistics, the source of the number series in this episode
- *Hollywood Handbook*, which Hayes co-hosts, and LA Material, where he works
- The AI provisions in the current WGA contract, worth reading against his "I have not heard of a show written by AI yet"
- *Ingrid Goes West*, which he calls a time capsule now
- Emily Sundberg's newsletter *Feed Me*, on life in New York

## The one thing to take away

One thing: **the decline usually starts before your chosen culprit shows up.** Shoot days peaked in 2016, then slid, collapsed, rebounded, and halved — and AI still hasn't written a show. The cost of naming a new technology as the only culprit is that you spend your time waiting for it to recede while the line that's actually falling goes unattended.

Here's something I've tried, and you can do it today. Take one thing you've complained about lately — work got harder, a group of friends drifted, the street you like went quiet — and pick a unit you can count (how many times a week people ask you for help, how many messages a month in that group chat, how many storefronts on that street are still open). Push it back five years and write down the year it started falling. Then compare that to the year your culprit appeared. If the two years are far apart, you have the wrong culprit.