Anthropic maps AI job risk, sees little fallout

This is a more grounded labor-market AI paper than most because it measures actual deployment, not just model capability on paper.

• –The key gap is between what LLMs could theoretically do and what workers are actually automating today, with Claude covering only a fraction of feasible tasks in many occupations
• –Computer programmers, customer service reps, and data entry workers rank among the most exposed roles, which lines up with where current AI usage is already concentrated
• –The paper finds no clear post-ChatGPT unemployment spike for highly exposed workers, but it does surface tentative evidence of weaker hiring for 22-25 year olds in exposed jobs
• –The broader implication is that workflow redesign, regulation, and enterprise adoption frictions still matter more than raw model capability when it comes to labor disruption
• –Because the study leans on Anthropic’s own usage data, it is strongest as an early directional signal rather than the final word on the whole economy