Anthropic just released labor market research on AI job displacement. An AI company publishing data about how AI affects employment. I have thoughts.
First, credit where it's due: most AI companies would never publish this. The incentive structure points entirely toward "AI creates jobs" messaging. Anthropic chose to publish data that makes their own product category look potentially disruptive to employment. That takes institutional courage.
What the data actually shows
The research isn't doomer clickbait. It's measured and specific. The findings suggest that AI is already displacing tasks in writing, coding, and customer service roles. Not entire jobs yet, but significant portions of what people do day-to-day.
The numbers I found most interesting: roles with heavy writing components are seeing 15-25% task displacement. Not layoffs. Task displacement. The person still has a job, but a quarter of what they used to do is now handled by AI.
That 15-25% number matters because it's in the uncomfortable middle zone. It's too much to ignore but not enough to justify eliminating the position. So what happens? Companies don't fire the writer. They just don't hire the next three writers they would have hired. The jobs don't disappear. They fail to appear.
Why this is hard to talk about
I build with AI daily. I'm a net beneficiary of this technology. And I still find the displacement data uncomfortable. That discomfort is worth sitting with.
The standard tech industry response to job displacement research is some version of "AI will create more jobs than it destroys, just like every previous technology." And historically, that's been true. The ATM didn't kill bank teller jobs. The spreadsheet didn't kill accountants.
But timing matters. The ATM transition happened over 20 years. AI task displacement is happening in 20 months. Even if the long-run outcome is net positive, the transition period can be brutal for real people with real mortgages.
What I think the honest position is
Here's where I land: AI is going to displace a significant number of tasks currently done by knowledge workers. Some of those workers will transition to higher-value work within their roles. Some will need to reskill. Some will have a genuinely difficult few years.
Pretending otherwise is dishonest. And I think dishonesty about this actually slows down healthy adaptation.
When tech leaders say "AI won't take your job," people don't prepare. When the data says "AI is already absorbing 20% of certain job functions," people can start planning. They can learn to work with AI tools. They can shift toward the parts of their role that AI handles poorly. They can make informed career decisions.
Honest data enables honest preparation.
What Anthropic gets right (and wrong)
The research methodology is solid. They're looking at actual task-level impact, not just vibes. They're measuring displacement across income levels, which reveals that this isn't just hitting low-wage workers. Mid-career professionals earning $60-100K are in the impact zone too.
Where I think the research falls short: it doesn't address the replacement quality question. AI can write a blog post. Can it write a blog post that changes how someone thinks? AI can answer a customer support ticket. Can it make an angry customer feel heard? The task displacement numbers don't capture the quality gap.
My ask to other founders
If you're building AI products, read this research. Don't dismiss it because it's inconvenient. Don't catastrophize it because doom sells. Just read it and sit with the numbers.
Then ask yourself: are you building something that makes humans more capable, or something that makes humans less necessary? Both are valid businesses. But only one of them lets you sleep well at night when the displacement data gets worse.
Anthropic published the data. The least we can do is take it seriously.