dead simple: (AI) automation gains can't be modeled via linear task-savings due to "the structure of bottlenecks and how automation reshapes worker time around them"

Nothing in the article about targeting the rate-limiting factors.

And on the first line of the first page, this gem of gratitude: "We thank Refine.ink, ChatGPT 5.2 Pro, and Claude Opus 4.5 for research assistance"

And on this government-sponsored paper, a warning that copying ANY portion of the text REQUIRES I accompany it with full credit, including the copyright notice, so that quote above puts me into noncompliance.