Stanford brief: no sign yet that AI is causing mass job losses, despite Amodei's apocalypse warnings
- Stanford's SIEPR policy brief finds AI's impact on aggregate employment is likely small right now, despite predictions like Anthropic CEO Dario Amodei's forecast of 50% white-collar job losses and 20% unemployment.
- Since 2022, unemployment for the most AI-exposed worker quintile rose 0.77 percentage points versus 0.85 points for the least-exposed group, suggesting a broadly softening labor market rather than AI-driven job cuts.
- Firms that adopted enterprise AI saw employment grow by 10% over the following two years, an effect concentrated among firms with the highest per-capita AI spending.
- Job postings for software developers, a highly AI-exposed occupation, grew faster than for other occupations over the past year, undercutting claims that AI is displacing coding jobs.
- The brief flags that a tough job market for recent graduates may be partly attributable to AI, even as overall employment effects remain hard to detect in aggregate data.
Hacker News 의견들
Coding agents like Claude Code and OpenAI Codex only really started working well in late November, and general agents like OpenClaw or ChatGPT Work came even later. Studies covering 2022 to end of 2025 might be missing the real capability jump entirely.
We're clearly in a transition period, so it's hard to draw steady-state conclusions from this data yet.
This isn't the first study to find this. Programmers and IT were massively overhired during the pandemic, that's the real driver of layoffs, and it's just easy to blame AI instead of admitting productivity gains aren't as big as hyped.
Any research on AI's impact is inherently a lagging indicator given how fast the field moves. Remember that paper claiming programmers were 20% slower with AI? Well...
Non-agent mode with GPT and Gemini still confidently gives wrong answers, but turning on agent mode fixed a lot of that for me.
Anecdotally I'm seeing way more recruiter interest than last year, but it tracks the hype cycle more than real demand. VCs are funding founders who promise to replace whole industries with agents, then hire people to manage the agents, while contacts outside startups still describe layoff threats and pressure to use AI tools.
This is exactly why AI labs can't just stop training and become profitable. They can't afford for studies like this to become credible, so they keep moving fast so nobody notices there's no clothes.
Companies have been using AI as a layoff excuse since well before 2026 according to layoffs.fyi. There's an uptick in 2026, but that could mean AI is really causing more layoffs, or just that it's become a more convenient excuse.
I got laid off from a big tech company at the end of January with 'AI' as the stated reason, but everyone on my team knew it was really 2022 overhiring. Got a new job by April without much effort, so I think the AI jobs apocalypse narrative is way overblown, more doom trolling from Anthropic and OpenAI's mouthpieces than reality.
Companies are also now dealing with rising API pricing, and my F100 clients are skeptical the productivity gains match the cost. Many are now demanding proof of value or they're cutting the AI spend entirely.
I poked around the job market after 18 months of solo building and it was brutal. Companies are asking for 4 years of 'agentic AI experience', they're just making up requirements without any real bar because they're buying the hype without understanding it.
Bogus requirement lists aren't new or AI-specific, tech HR has been asking for X years of a technology that hasn't existed that long for as long as I can remember.
I'd avoid 'AI-native' companies entirely, they seem to attract burnout culture and delusional expectations.
Job listings aren't literal requirements, they're a wishlist and a rough proxy for skill level. If you can demonstrate the skill, missing a year or two of 'required' experience usually doesn't matter.