organisation
A three-month randomised field experiment examined AI-assisted patent drafting among 133 lawyers at eleven US firms. The NBER working paper, Does AI Assistance Enhance or Erode Expertise?, reports improved drafting quality with AI, with larger gains among junior lawyers.
At the end of the experiment, participants reviewed an existing patent application without AI. Previously assisted lawyers performed better on average, but the advantage was concentrated among senior lawyers. Junior lawyers showed no average improvement, with more dispersed results. This does not mean that every junior learned nothing. Nor were participants tested after spending three months without the tool.
Our perspective
Better assisted output and lasting skill development are different outcomes. Our recommendation is to preserve some unaided practice for junior employees and assess learning separately from deliverable quality. This is a lesson to test in your own functions, rather than a conclusion that every profession will reproduce the patent-law results.
The study concerns patent work in one setting and is a working paper. It was funded by Google and tested its InFlow tool; six of the seven authors are reported as Google employees or contractors. These relationships and the detailed protocol must be read alongside the results.
Sources
David Autor and co-authors, Does AI Assistance Enhance or Erode Expertise? Evidence from a Three-Month Field Experiment in Patent Drafting, NBER working paper 35720, September 2026, DOI 10.3386/w35720.
Explore further
Le Signal Spentia
Three dated, sourced facts, an article and a podcast. Delivered in French every Monday.