A preregistered randomized trial involving 133 patent attorneys at 11 US firms found that AI-assisted drafting quality was 0.38 standard deviations above the control group after 90 days. Blinded experts assessed the work, and immediate drafting gains were larger for junior attorneys. However, improvement on an unaided editing task after three months was concentrated among senior attorneys; juniors showed no average gain. That does not mean every junior attorney lost ability. Given the short duration and limited sample, organizations should assess output quality separately from the development of independent expertise.