AI 为什么还没推高失业率：增强、专业知识与初级招聘警讯

Current United States labor data shows no material AI-driven increase in unemployment, including among highly exposed occupations.

So far, AI has behaved like a skill-biased labor-augmenting technology that raises the returns to complementary expertise.

Jobs are not fixed bundles of tasks, so automating many tasks can expand a role's scope instead of eliminating the role.

Essential tasks that remain difficult to automate can constrain end-to-end automation and preserve the value of human labor.

Claude Code users with deeper domain expertise complete more valuable tasks, succeed more often, and recover from failures more effectively.

AI's labor impact may appear first as reduced hiring for junior workers rather than visible waves of layoffs.

The labor effects of AI will be uneven because some occupations contain automatable core tasks while others depend on coordination, presence, and judgment.

Faster productivity growth can coexist with stable employment when AI expands output, responsibilities, and demand instead of only cutting headcount.

Model routing is becoming a first-class infrastructure layer as companies optimize quality, cost, latency, and provider dependence.

Smaller models tuned inside a real product harness can approach frontier performance on bounded tasks at substantially lower cost.