How did Anthropic measure AI's "theoretical capabilities" in the job market?

Arstechnica examines the methodology behind Anthropic's claims about LLM job displacement impact, highlighting that the underlying research (Eloundou et al) relies on unbounded adoption timelines and forward-looking assumptions about future software capabilities. The article shows a significant gap between "theoretical capabilities" (47-56% impact) and practical adoption, noting that most real-world use would require widespread buy-in that researchers themselves acknowledged is uncertain.