FULL STORY

Autor's RCT: AI Boosts Patent Lawyers' Output

An NBER randomized trial by David Autor's team found AI drafting help significantly improved patent lawyers' quality and productivity, with senior attorneys gaining most, while junior lawyers' gains vanished once the tool was removed.

2026-09-19 ~ 2026-09-24 · 2 episodes · 7 posts

Episode 1 · NBER trial: AI drafting tool benefits senior patent lawyers most (2026-09-19, 2 posts)

A preregistered three-month RCT by MIT economist David Autor's team and Google gave a custom AI drafting assistant to 133 patent lawyers at 11 US IP firms. Blinded expert ratings showed the largest gains for senior lawyers, with mixed results for novices.

Episode 2 · NBER RCT: AI Boosts Junior Patent Lawyers' Output but Leaves No Lasting Skill Gains (2026-09-22, 5 posts)

David Autor's team published a three-month preregistered RCT (NBER Working Paper 35720) examining AI assistance in patent drafting. 133 practicing patent lawyers across 11 US IP firms received a custom AI drafting assistant, with output scored by blinded expert attorneys. AI improved output quality, but gains were sharply stratified: juniors' improvements vanished once the tool was removed, while seniors' gains persisted.

Confirmed

  • Design: preregistered RCT, 11 US IP firms, 133 practicing patent lawyers, three months, custom AI drafting assistant, blinded expert grading (m2, m5).
  • With AI, patent documents were better at both 10-day and 90-day checkpoints; juniors benefited most, gaining 0.58 and 0.60 standard deviations respectively (m1, m2).
  • After tool removal, junior gains disappeared entirely, leaving no durable capability or skill improvement (m3, m4).
  • Senior gains persisted after tool removal; @zivravid infers that solid underlying judgment/experience explains the difference (m4).
  • AI assistance did not improve juniors' underlying judgment (m1, m5).

Why it matters

  • @zivravid notes the contrast with earlier Stanford employment data: even juniors who keep their jobs may see skill accumulation stall due to AI reliance, raising concerns about weakened early-career skill formation.
  • For organizations, the experiment suggests layered AI value: it is a productivity amplifier for juniors, not a teacher, and talent-development strategies may need redesign.