The headcount debate gets the attention and it is the less tractable question. This piece is useful because it goes to the one that firms can actually act on: junior roles are changing first and most, shifting away from producing first drafts, research memoranda and document review toward reviewing, validating and overseeing output produced by a system.
The difficulty is that the displaced tasks were not merely output — they were the training. Document review taught what matters in a document. Research memos taught the map of an area, including the wrong turns, which is where most of the learning was. First drafts taught how a document is constructed and why each clause is there. Remove them and the firm still gets the work product, and stops getting the lawyer.
The circularity is the part to sit with. The plan most firms describe is that AI produces and juniors verify — but verification presupposes enough knowledge to know when something is wrong, and that knowledge came from doing the work AI now does. A junior who has never built the analysis can confirm a citation exists and cannot judge whether the reasoning holds. Firms that have not addressed this have a plan that assumes its own output.
Whether firms will need fewer juniors is genuinely contested and both positions are coherent: one holds that automating junior work reduces the need for juniors, the other that cheaper legal work expands demand and that supervising AI at scale still takes people. What is much less contested is that the composition of junior work is changing sharply — which affects training regardless of how the headcount question resolves, and is the reason this entry is filed under development rather than economics.