Is AI actually helping?

TL;DR

• 1

  The white-collar bloodbath has not shown up in the data: Berman points to Sam Altman, Dario Amodei, Goldman Sachs CEO David Solomon, Apollo Research, and ADP employment data to argue there is still "zero evidence" of broad AI-driven job losses.

• 2

  A lot of AI layoffs look more like cleanup from overhiring: He cites Duolingo, Pinterest, Amazon, Meta, Block, and Twitter as examples where companies blamed AI while likely correcting for bloated headcount from the zero-interest-rate era.

• 3

  Jevons paradox is his core explanation for why AI can raise jobs, not kill them: As AI gets cheaper and more useful, companies attempt more projects, spend more overall, and still need humans to prompt, verify, package, market, and sell the output.

• 4

  The cost problem is real, but bad AI usage is an even bigger problem: Uber reportedly burned its entire 2026 AI budget in four months, and Berman says many firms overspend on frontier models like Claude Opus when cheaper options such as DeepSeek could handle many use cases.

• 5

  Only a tiny group really knows how to use AI at the frontier: He highlights Peter Steinberger spending about $1.3 million in tokens to build a "software factory," not just generate code, as an example of advanced AI use that most enterprises cannot yet copy.

• 6

  The gap is expectation versus reality, not value versus no value: Berman says startups like Pulsa and some YC-style AI-native company rhetoric are selling a future state as if it already exists today, when end-to-end autonomous companies still need substantial human guidance and verification.