200,000 Fewer Creative Jobs in the US in 2026: Is AI to Blame?
Hello HaWkers, at the end of September economist Joseph Politano, from the Apricitas Economics newsletter, published an analysis that took over Hacker News: the United States lost more than 200,000 jobs in the creative industries in four years, around 50,000 in the last twelve months alone. Film and TV by themselves lost more than 100,000 positions, almost a third of the sector.
The question everyone asked next was the obvious one: is this artificial intelligence arriving? And if AI is emptying studios, newsrooms, and publishers, how long until it does the same to the work of people who write code?
In this article I'll separate what the data actually says from what the headlines say, show how you can check these numbers yourself with a few lines of Python, and pull out the practical lessons that apply to any developer's career in 2026.
What the Numbers Show
Politano's analysis, published on September 27, 2026 under the title "AI and the Fall? of the Creative Class", relies on public employment data from the US government. The highlights:
- More than 200,000 jobs lost in the creative industries over four years, with the recent pace close to 50,000 per year.
- Film and TV lost more than 100,000 positions, almost a third of the sector's entire workforce.
- Written publishing (newspapers, magazines, books) lost more than 70,000 positions in the same period.
- The intensity and duration of the decline are only matched by the 2001 and 2008 recessions. The difference is that today there is no broad recession.
A few days later, on October 2, Fortune ran its own check against the Bureau of Labor Statistics (BLS) series and arrived at similar numbers:
- The motion picture and sound recording sector fell 27%, the equivalent of 120,000 jobs, between August 2022 and August 2026.
- In June, that sector had 322,300 jobs, the lowest level since 1995 if we set the pandemic aside.
- Shoot days in Los Angeles fell from 36,792 in 2022 to 19,694 in 2025.
- Meanwhile, performing arts and live spectator sports gained around 67,000 positions over the same four years.
That last point is the most interesting of all, and I'll come back to it later.
Is AI Really to Blame?
Here is the part almost no headline repeated: Politano himself says it is not possible to isolate how much of this loss is caused by AI. In his words, "proving the exact amount of AI-driven job loss in the arts is extremely difficult".
There are at least four forces acting at the same time:
- Consolidation in Hollywood. Mergers between studios and platforms reduce productions and cut duplicated structures.
- Production moving out of the US. Shooting in Canada, the UK, or Eastern Europe is cheaper, and the Los Angeles numbers show it.
- Independent creators capturing attention. YouTube, Twitch, and TikTok compete for the same screen time as cable TV, and the transition to streaming has been, according to Politano, "extremely difficult" for the studios.
- The 2023 strikes. The BLS itself noted that the work stoppages that year helped kick off a downward trend that stretched into 2026.
And the decline in written publishing started long before ChatGPT. American newspapers once employed around half a million people, more than the entire oil industry, and today they are close to 85% below the peak. Magazines have cut 66% of their staff since the 1990s.
One data point helps calibrate expectations: according to the St. Louis Federal Reserve series cited in the article, about 32% of workers in arts, entertainment, and recreation use generative AI to some degree, against a much higher overall average across all sectors. In other words: the sector that lost the most jobs is not the one that uses AI the most.
The counterpoint from the person who coined the term
Fortune spoke with Richard Florida, the urban theorist who coined the expression "creative class" in his 2002 book. His answer was blunt: "I'm not sure his numbers are right". Florida believes AI will not eliminate these jobs, but rather change what these people do, putting more value on a specific kind of work.
He splits work into three groups: manual, cognitive, and social/human/entrepreneurial. AI automates the first two and raises the value of the third. The most exposed jobs, according to him, are the "grind-out" ones: entry-level roles built on a lot of repetitive mental effort, like the long hours of junior analysts at consulting firms and banks.
How to Check the Data Yourself
One of the best things about this discussion is that the data is public. Fortune cited the FRED series (the economic database of the Federal Reserve Bank of St. Louis), and anyone can download them as CSV without an API key.
The script below downloads the CES5051200001 series (employment in motion pictures and sound recording, in thousands) and calculates the drop since the peak:
# Calculates the drop in motion picture and sound recording employment since the peak
# Source: FRED (Federal Reserve Bank of St. Louis), series CES5051200001
import pandas as pd
SERIE = "CES5051200001"
URL = f"https://fred.stlouisfed.org/graph/fredgraph.csv?id={SERIE}"
# The CSV comes with two columns: the date and the series value
df = pd.read_csv(URL)
df.columns = ["data", "empregos_mil"]
df["data"] = pd.to_datetime(df["data"])
# Missing values show up as "." in the CSV, so we convert them safely
df["empregos_mil"] = pd.to_numeric(df["empregos_mil"], errors="coerce")
df = df.dropna()
# Only consider the period from 2022 onwards to find the recent peak
recente = df[df["data"] >= "2022-01-01"]
pico = recente.loc[recente["empregos_mil"].idxmax()]
atual = df.iloc[-1]
queda_abs = pico["empregos_mil"] - atual["empregos_mil"]
queda_pct = queda_abs / pico["empregos_mil"] * 100
print(f"Pico: {pico['data']:%m/%Y} com {pico['empregos_mil']:.1f} mil empregos")
print(f"Atual: {atual['data']:%m/%Y} com {atual['empregos_mil']:.1f} mil empregos")
print(f"Queda: {queda_abs:.1f} mil empregos ({queda_pct:.1f}%)")Running this lets you see with your own eyes the order of magnitude Fortune reported. It's the kind of check that takes five minutes and keeps you from repeating a number on LinkedIn that doesn't hold up.
The Detail That Changes the Reading: Digital Falls, In-Person Grows
The most revealing data point in the analysis is not the total loss. It's the pattern: every digital subsector of the arts is shrinking, while in-person entertainment is growing. Museums, performing arts, and live shows keep hiring.
You can compare the two series side by side with the same method:
# Compares a digital sector (film and sound) with an in-person sector
# (performing arts and spectator sports) using FRED series
import pandas as pd
SERIES = {
"CES5051200001": "Cinema e gravação de som",
"CES7071100001": "Artes cênicas e espetáculos",
}
def baixar(serie: str) -> pd.Series:
"""Downloads a FRED series and returns numeric values indexed by date."""
url = f"https://fred.stlouisfed.org/graph/fredgraph.csv?id={serie}"
df = pd.read_csv(url)
df.columns = ["data", "valor"]
df["data"] = pd.to_datetime(df["data"])
df["valor"] = pd.to_numeric(df["valor"], errors="coerce")
return df.dropna().set_index("data")["valor"]
for serie, nome in SERIES.items():
dados = baixar(serie)
# Uses August 2022 as the baseline, the same starting point Fortune used
base = dados.loc["2022-08-01"]
atual = dados.iloc[-1]
variacao = (atual - base) / base * 100
print(f"{nome}: {base:.1f} mil -> {atual:.1f} mil ({variacao:+.1f}%)")Politano's reading of this is very useful: music adapted to the internet by moving revenue to live shows. Video has no equivalent place to escape to. The business model of film and TV, according to him, is "fundamentally being squeezed".
Notice that this is a story about a business model, not about a specific tool. And that is exactly where the lesson for people who work in software lies.
And Where Does Tech Fit Into All This?
Politano draws an explicit bridge between the arts and technology. In 2022, the US tech sector created almost 300,000 jobs in a single year. Since then, it has lost positions every year, in the longest slowdown in the sector's recent history.
His conclusion is cautious: outside a few narrow sectors, AI-driven job displacement is still relatively small. But the arts may be a warning sign for neighboring kinds of work, and tech is one of them.
If you follow the blog, you've already seen how this shows up in entry-level positions. I wrote about it in the junior developer crisis in 2026 and how to stand out, and the pattern is similar: it's not that the work vanished all at once, it's that the entrance door got narrower.
Another parallel is worth remembering. On March 24, 2026, OpenAI announced it was shutting down the Sora app, the video generator that had scared Hollywood just a few months earlier. The tool that looked like the biggest direct threat to film and TV left the stage, and jobs kept falling. Again: the bigger force seems to be economic, not a single tool.
A Practical Exercise: Map Your Own Exposure
Richard Florida's argument gives you a simple framework to look at your own routine. If AI automates manual and repetitive cognitive work first, it's worth measuring how much of your week falls into each group.
The script below is a personal tool, not a scientific study. You list your tasks, estimate hours per week, and classify each one:
// Personal exposure map based on Richard Florida's three types of work
// manual: repetitive execution | cognitive: standardized analysis | human: judgment, taste, people
const tarefas = [
{ nome: 'Escrever CRUD e boilerplate', horas: 8, tipo: 'manual' },
{ nome: 'Corrigir bugs reportados com stack trace', horas: 6, tipo: 'cognitivo' },
{ nome: 'Escrever testes unitários de funções simples', horas: 4, tipo: 'manual' },
{ nome: 'Definir arquitetura com o time', horas: 4, tipo: 'humano' },
{ nome: 'Conversar com usuários e produto', horas: 3, tipo: 'humano' },
{ nome: 'Revisar pull requests de colegas', horas: 5, tipo: 'cognitivo' },
{ nome: 'Mentorar pessoas juniores', horas: 2, tipo: 'humano' },
];
// Approximate automation exposure weight per type of work
const exposicao = { manual: 0.8, cognitivo: 0.5, humano: 0.15 };
const totalHoras = tarefas.reduce((soma, t) => soma + t.horas, 0);
// Sums hours per type to show where your week is concentrated
const porTipo = tarefas.reduce((acc, t) => {
acc[t.tipo] = (acc[t.tipo] || 0) + t.horas;
return acc;
}, {});
const indice =
tarefas.reduce((soma, t) => soma + t.horas * exposicao[t.tipo], 0) / totalHoras;
console.log('Horas por tipo:', porTipo);
console.log(`Índice de exposição da sua semana: ${(indice * 100).toFixed(0)}%`);
// The most exposed tasks are the first candidates to delegate to AI
const maisExpostas = [...tarefas]
.sort((a, b) => exposicao[b.tipo] * b.horas - exposicao[a.tipo] * a.horas)
.slice(0, 3)
.map((t) => t.nome);
console.log('Comece delegando:', maisExpostas.join(', '));The weights are arbitrary on purpose: the value of the exercise lies in listing your tasks honestly. If more than half of your week is work that a coding agent already does reasonably well, that doesn't mean you'll lose your job tomorrow. It means those hours are the first to get cheaper, and that your value needs to move toward the others.
What to Do With This Information
Putting Politano's data together with Florida's reading, you can draw five concrete lessons for a career in tech:
1. Be suspicious of single-cause headlines
"AI destroyed 200,000 jobs" is a sentence the author of the study never wrote. Consolidation, offshoring, changing audience habits, and strikes explain a good chunk of the decline. When you read a number like that about the software field, look for the source and check it.
2. Look at your employer's business model
Film and TV are losing jobs because the money moved somewhere else. Ask yourself: does the company you work for make money in a way that is growing or shrinking? That answer predicts more about your stability than any AI model benchmark.
3. Invest in work that isn't "grind-out"
Florida mentions judgment, taste, originality, imagination, asking good questions, and mobilizing people. In software, that becomes: understanding the user's problem, deciding what not to build, critically reviewing the code AI generated, and aligning the team.
4. Use AI for repetitive work, not against yourself
People who learn to delegate boilerplate, simple tests, and migrations to coding agents gain hours for the work that's worth more. People who ignore the tool keep competing exactly in the band that's getting cheaper the fastest.
5. Value what happens "live"
The fact that in-person entertainment grew while digital fell has a dev version: community, events, mentoring, client relationships. Things that depend on presence and trust are hard to automate.
Outlook for the Coming Years
Politano himself admits that predicting the effect of AI is hard. He cites BookTok as an example of a surprise: nobody predicted that a short-video social network would help sell books, and book sales rose about 70% in six years.
His view is that, across the economy as a whole, the biggest impact of AI-generated art will be cultural and social, not macroeconomic, because the arts are a small slice of GDP. But he makes an important caveat: if the economic impact of AI gets anywhere close to the trillions of dollars invested in it, the transformation will go far beyond these sectors.
For people who work in tech, the healthiest stance sits halfway between panic and denial. The data shows that the American creative market is going through one of the worst phases in its recent history, and it also shows that AI is not the only explanation, nor, for now, the main one. Following the numbers closely, understanding the business model of where you work, and moving your time toward work that requires human judgment are choices that pay off with or without an AI revolution.
Let's go! 🦅
📚 Want to Keep Up With What Is Coming?
This article covered the drop in creative jobs in the US and what it teaches about a dev career, but the ecosystem changes every week and not everything becomes an article here.
On X I share what I'm testing, behind the scenes of my projects, and the news that shows up before it turns into a post.
Follow Me There
💡 Daily content about development, career, and the tools I actually use