AI assistance improved patent drafts in a three-month randomized trial involving 133 attorneys at 11 US intellectual-property firms, with the largest immediate gains among lawyers with less than seven years of experience. The lasting benefit was much less clear.
Two-thirds of participants received InFlow, an unreleased Google Labs assistant, for two patent-claim drafting tasks. The remaining attorneys worked without AI. Independent reviewers scored the work for enforceability, technical accuracy, strategic ambiguity, completeness and clarity. AI users performed better after both 10 and 90 days, largely by producing fewer poor drafts rather than more top-rated ones.
A third task tested unaided judgment: everyone had to review and mark up an existing patent draft without AI. Senior attorneys who had used InFlow performed better, but junior users showed no consistent improvement despite gaining most during assisted drafting. Their results split between stronger and weaker outcomes.
The Google-funded working paper does not prove AI caused skill loss. Its sample was small, it lacked a baseline skills test and three months is short beside a legal career. Still, the contrast suggests that productivity and learning are not the same outcome. Firms adopting assistants may need deliberate unaided practice and mentorship if junior staff are to build judgment rather than only produce better work while the tool is present.