Artificial Intelligence Boosts Immediate Work Quality but Fails to Build Lasting Skills in Junior Lawyers
Cambridge, Thursday, 24 September 2026.
A new study reveals artificial intelligence improves legal work quality, but junior lawyers lose all performance gains without the tool. Lasting expertise requires baseline knowledge before adopting automated assistants.
NBER Field Experiment Quantifies AI Impact on Legal Drafting
A recent three-month field experiment published by the National Bureau of Economic Research (NBER) provides empirical evidence on the dual impact of artificial intelligence assistance on high-skilled professionals [1]. Conducted with 133 practicing patent lawyers across eleven U.S. intellectual property law firms, the study measured performance while using a custom AI drafting assistant and professional judgment afterward without it [1][2]. The research highlights a critical tradeoff for corporate leaders and policymakers evaluating enterprise AI adoption between immediate operational efficiency gains and the potential degradation of specialized human capital over time [1]. All work was scored by blinded expert patent attorneys to ensure objective assessment of the output quality [2].
Immediate Output Quality vs. Long-Term Skill Acquisition
Parallel to findings from other white-collar domains, AI access raised the quality of work delivered on benchmark patent drafting tasks significantly during the trial period [1]. At 10 days, the quality score increased by 0.34 standard deviations, and by 90 days, the increase reached 0.38 standard deviations, representing a marginal gain of 0.04 SD over the initial period [1]. Junior lawyers experienced the largest gains in output quality while using the tool, indicating that AI assistance effectively boosts short-term productivity for less experienced practitioners [2]. However, the study notes that whether AI assistance builds or erodes professional expertise remains a complex question requiring analysis of performance after the tool is removed [1].
Divergence in Skill Retention
After three months, all subjects were tasked to redline an existing patent application without AI, a core task of patent practice requiring expert judgment [1]. Treated lawyers outperformed controls by 0.32 standard deviations overall, but this advantage was concentrated entirely among senior lawyers who scored 0.45 standard deviations higher than peers [1]. Junior lawyers showed no average gain in this unassisted task, suggesting that foundational expertise may be a prerequisite for extracting durable skill from AI-assisted practice [1][4]. This indicates that while AI enhances output, it does not necessarily transfer knowledge to the user without a baseline of expertise.
Performance Bifurcation Among Junior Lawyers
The performance scores for junior lawyers bifurcated significantly when the AI tool was removed, with sharply fewer mediocre scores offset by more poor and more good ones [1]. This distribution suggests that without the tool, junior performance became more volatile rather than consistently improved [3]. Experts note that even juniors who keep their jobs may be losing ground in terms of durable skill acquisition, as gains vanished with zero durable improvement once the tool was taken away [3]. The largest gains from AI thus accrued to the lawyers who retained the least independent capability without it [1].
Economic Implications for Legal Markets
For the broader economy, these findings suggest a potential risk where reliance on AI could erode the pipeline of expert legal talent required for complex judgment tasks [2]. Firms are advised to deploy AI so that it builds judgment instead of replacing it, ensuring attorneys continue doing substantive work like reviewing and restructuring tool outputs [2]. The study was funded by Google and remains a working paper not yet peer-reviewed, indicating that further validation may be required before definitive policy shifts [2]. Nonetheless, the data underscores the need for active stewardship rather than passive adoption of AI in professional services [4].
Strategic Recommendations for Enterprise Adoption
Corporate leaders must balance the immediate efficiency benefits of AI with long-term human capital development strategies to prevent expertise erosion [1]. The lesson for firms is not to hold AI back from associates but to structure workflows that mandate human verification and substantive engagement with AI-generated content [2]. As of September 2026, this data provides a crucial framework for understanding the economic trade-offs of generative AI in high-skill sectors [3]. Future research will need to monitor whether intervention strategies can mitigate the observed skill atrophy among junior professionals [4].