David Autor, lead author of a patent-law AI trial, says employers should preserve some work that junior lawyers do without assistance. In a new interview, he suggested that beginners frame an argument themselves before asking an AI assistant to challenge its logic, while warning that this workflow has not been tested.
The recommendation adds a workplace question to an experiment that measured both AI-assisted drafting and a separate task completed without AI.
The randomized trial gave two-thirds of 133 lawyers at 11 U.S. intellectual-property firms early access to a Google Labs patent-drafting assistant. After 90 days, the access group scored 0.38 standard deviations higher than controls on a simulated patent-drafting task. The authors say that gap corresponds to an 11-point rise in percentile ranking among control-group scores. The test’s scores were based on five dimensions, including accuracy, enforceability and clarity.
At the 90-day mark, 91 of the 133 lawyers completed a separate test without AI by marking up a hypothetical patent with substantive and stylistic errors. Patent professionals graded the submissions. Lawyers with at least seven years of experience in the AI-access group scored 0.45 standard deviations above senior controls, which the authors equate to a 13-point rise in percentile ranking. Compared with junior controls, junior lawyers’ scores included more poor and more good results, without an increase in excellent scores.
Autor told Fortune that newer lawyers could first set out an argument on their own, then use AI to pressure-test its logic. He also warned employers against removing formative tasks without providing another way to learn. These are his recommendations, not practices tested by the trial.
The study was limited to 133 patent lawyers at 11 firms and followed them for three months. Its authors say those results do not establish how AI affects expertise over a full career. Fortune reported that the NBER working paper had not been peer-reviewed. Google Research says the 11 participating firms had regular, non-exclusive business relationships with Google. Google conducted the study and paid its direct costs; the paper lists five co-authors as full-time Google employees and Josh Martin as a paid Google contractor.
Sources
- Fortune, “The economist behind the ‘China shock’ says AI is a paradoxical ‘skill-disequalizer’ for Gen Z workers,” Oct. 10, 2026 (interview with David Autor)
- Google Research, “Does better work always mean better workers?”, Oct. 7, 2026
- National Bureau of Economic Research, Working Paper 35720, “Does AI Assistance Enhance or Erode Expertise?”, September 2026, revised October 2026
- Illustrative stock photograph: Pavel Danilyuk, Professional Lawyer writing on a Notebook, Pexels (Pexels License).