The economist behind the ‘China shock’ says AI is a paradoxical  ‘skill-disequalizer’ for Gen Z workers

A three-month, 133-lawyer experiment at 11 U.S. intellectual-property firms found that AI access raised patent-drafting scores by 0.38 standard deviations (described by the authors as an 11-percentile-point gain), but on an AI-free test, only senior lawyers—defined as at least seven years in practice—showed an average improvement: 0.45 standard deviations versus control peers; juniors showed no average gain. The NBER working paper is not peer-reviewed, only 91 lawyers completed the unaided test, and the authors say the study does not establish whether AI prevents juniors from becoming experts over a career. Autor warned that automating formative practice without replacing it with guided learning could weaken the apprenticeship pipeline, while emphasizing that human and machine intelligence can complement one another.

Analysis

The investable signal is not “AI replaces lawyers”; it is that output quality can rise before underlying judgment does. If firms respond by shrinking junior cohorts, they may book near-term labor savings while quietly degrading the future supply of experienced reviewers. That creates a delayed quality-control and liability cost, especially in work where fluent errors are hard to detect. The effect could extend beyond patent drafting to other professional services, but this small, three-month study does not establish a durable career-long outcome.

For Alphabet (GOOG), the study is modest evidence that AI can improve workflow output, not evidence of paid adoption, pricing power, or a defensible product advantage. Google’s role in funding and conducting the work also argues for discounting promotional interpretation. Established legal-information and workflow vendors such as Thomson Reuters and RELX could benefit if buyers prioritize verification, domain content, and review controls over generic drafting; that is a hypothesis, not a demonstrated share shift.

Over days, this is unlikely to justify a material re-rating. Over 1–3 months, watch law-firm hiring, AI workflow deployments, and product disclosures for evidence that productivity gains convert into lower costs or vendor revenue. Over 6–18 months, the key structural test is whether employers preserve supervised training and independent skill checks. A contrarian risk to the cautious read: better tools or deliberate AI-assisted training may let juniors acquire judgment faster than this trial detected. The thesis weakens if later, larger studies show durable junior skill gains, or if firms demonstrate lower review/error rates without reducing apprenticeship.

Key Decisions for Investors

• No standalone GOOG trade: treat the paper as a weak adoption signal, not proof of monetizable demand or competitive advantage. Revisit if Alphabet discloses enterprise AI usage translating into paid seats, retention, or material revenue contribution.
• Track legal-services hiring and workflow economics over the next 1–3 months. A sustained reduction in junior hiring alongside stable matter quality would support labor-substitution exposure; rising rework, errors, or partner-review burden would challenge it.
• Monitor Thomson Reuters and RELX disclosures for demand tied to domain-specific content, verification, or AI-assisted legal workflows. Prefer evidence of paid usage and customer retention over product launches or vendor claims.
• For a 6–18 month thesis check, seek independent, larger-sample evidence on unassisted junior performance and error rates. Durable junior skill gains would falsify the apprenticeship-risk view; output gains without unaided competence would strengthen it.