The company building the AI tools supposed to automate your job just admitted they haven't actually automated any jobs yet.

The Summary

The Signal

Peter McCrory's internal analysis lands like a red flag in the middle of the hype cycle. Anthropic's own economist looked at the U.S. labor data and found nothing. No mass layoffs in knowledge work. No crater where customer service reps used to be. No white-collar apocalypse.

This matters because Dario Amodei has been one of the more credible voices warning about AI's impact on professional work. When your CEO says the models will replace lawyers and analysts and your economist says the data shows none of that happening, someone's reading the tea leaves wrong.

"We don't see significant impact of AI on the U.S. labor market."

Two explanations fit the data. First: the models aren't actually good enough yet to replace human judgment at scale, despite the benchmarks and demos. Second: they are good enough, but companies haven't figured out deployment. The tech is ready but the org charts, workflows, and trust aren't.

McCrory's timing suggests he believes the second explanation. You don't publish this kind of analysis unless you're trying to explain a lag, not a permanent gap. The models work in the lab. They don't work yet in the Fortune 500 because enterprise adoption is a culture problem disguised as a procurement problem.

Key points from the data:

  • Labor force participation remains stable across knowledge work sectors
  • No statistical employment shift in roles most exposed to AI capability
  • The delay between capability demonstration and labor market impact is widening

The Implication

If McCrory is right about the lag, this is the calm before the actual disruption. Companies are still in pilot mode. Legal is still drafting AI usage policies. Compliance is still figuring out liability. That window closes when the first major firm reports a 30% headcount reduction with flat or growing revenue.

For workers: the absence of impact today doesn't mean protection tomorrow. It means you have time to build skills the models can't replicate yet. For companies building AI tools: the deployment problem is now the product problem. Whoever solves enterprise trust and workflow integration wins the actual market, not just the benchmark leaderboard.

Sources

Fortune Tech