AI is cutting junior associates' work; some law schools are already redesigning how they train lawyers

By Zendoric · October 5, 2026.

For decades, a junior lawyer learned the trade by doing repetitive tasks —reviewing documents, drafting, researching case law— under the almost always distracted gaze of a partner in a hurry. That implicit contract is breaking down: AI already does much of that routine work, and clients are no longer willing to pay for the hours a junior needed to learn on the job. That is the starting point of a panel held at RelFest (formerly Relativity Fest), as recounted by lawyer and blogger Stephen Embry in Above the Law: if firms no longer have —or pay for— the tasks that used to train lawyers, someone has to take on that training. And the natural candidates are law schools.

The problem is that most of those schools have been operating the same way for more than fifty years: Socratic method, casebooks, a single five-hour exam at the end of the course. It is the model portrayed in "The Paper Chase", the 1973 film about the feared Professor Kingsfield. According to the article itself, some schools, far from adapting, outright ban their students from using AI.

Against that inertia, the panel —with April Dawson (Howard), Dan Linna (Northwestern Pritzker), Bobby Williams (iDS/NExT) and Dorna Moini (Relativity/Gavel)— presented something different: not laments about how AI degrades legal education, but concrete tools to accelerate it. Linna has built a Socratic AI tutor that questions each student almost daily, instead of the couple of times per semester an in-person class allows; he also uses automated coaches that give feedback on submitted work, an oral argument simulator and a full trial preparation simulation. Dawson, who teaches administrative law, asks her students to read with an AI assistant alongside them —to ask it what a term means or what the judge meant in a given quote— and then complete a Socratic questionnaire generated by the tool itself, which she grades. In another of her classes she has the AI build a profile of a self-represented litigant ("pro se") and asks students to navigate the court system with the Courtroom5 tool while in that role, so they understand how it feels for someone without legal representation. Williams, for his part, insists that law school must also teach project management and human relationships —he even uses theatrical improvisation— because, as he puts it himself, "life is project management". And Moini recalls the trade's most uncomfortable lesson: however much you prepare an argument, you do not know how well you command the material until the judge throws you the question you had not anticipated.

None of this is free. Linna acknowledges that it requires changing how professors assess —more frequent checks, less reliance on the single annual exam— and, above all, giving up a good dose of academic ego: the article recounts the anecdote of a law school that rejected an innovation proposal on the grounds that "if it were a good idea, we would have thought of it already". That resistance, more than the technology itself, is the real obstacle.

Our take is that this episode, however niche it may seem —a panel at a tech trade show for lawyers—, is an early case study of a problem that will repeat itself in every profession that trains through learning by repetition. We have already analyzed how AI narrows the base of the pyramid at law firms: the mechanical work that used to train people disappears, and expert judgment and the client relationship rise in value. What this panel adds is the other half of the equation: if entry-level work disappears as a training path, the practice hours once obtained by accumulating real tasks have to be deliberately manufactured. And that applies equally to medical residency, the journalist who learned by editing minor pieces, or the junior banking analyst. The question is not only "what work does AI destroy", but "what training mechanism does it destroy along the way, and who replaces it".

In the short term, the picture is uneven and there is no point hiding it: most law schools are not moving, some flatly block AI, and meanwhile firms keep cutting junior tasks. That opens a quality gap between graduates of schools that are redesigning their teaching —Northwestern, Howard, and according to the article also Vanderbilt and Suffolk— and those from the rest, who may come out with fewer real hours of deliberate practice than previous generations. It is a starting inequality that will add to the one we already see in the legal job market.

But the underlying horizon is encouraging, and it connects with the thesis we hold at Zendoric: used well, AI does not only replace tasks, it also democratizes something only large firms with mentors on hand could previously afford —constant feedback, repeated practice, realistic courtroom simulations. A Socratic tutor that is there almost every day, instead of a couple of times a semester, is the kind of personalized attention that used to depend on having a partner generous with their time, a scarce commodity at any firm. If the experiment takes hold, the result is not "zombie AI lawyers" churning out automated garbage, as the article itself fears, but graduates with the judgment of a third- or fourth-year associate on their first day of work. That is no small promise: it is the difference between a profession that shrinks from the bottom without replacing anything, and one that uses the same technology that took away its routine work to shorten the path to expert judgment, which is, in the end, the only thing a client really pays for.