The most widely read magazine for Canadian lawyers
Issue link: https://digital.canadianlawyermag.com/i/1544922
32 www.canadianlawyermag.com OPINION UPFRONT AI risks hollowing out apprenticeship model As I enter the profession, my generation could have the most powerful legal tools in history, with no ability to assess when to trust them Arjun Gupta is a JD Candidate 2026 at the University of Ottawa, Faculty of Law. You can see some of his work at www.arjungupta.com. Reviewing a contract AI has already flagged is a fundamentally different cognitive act from reviewing one cold. The first tests recognition; the second builds it. Westlaw changed how lawyers find cases, but they still must read them and reach their own conclusion. AI proposes the conclusion for you. Let me be direct: articling students should be using AI as much as possible. This is a paradigm shift. AI can coach you through a memo, flag the errors you made on your last analysis, and surface patterns across a thousand documents no human could hold in working memory. The senior partner of 2036 will not resemble the senior partner of today, and training articling students as though she will is, respectfully, a mistake. I remember watching taxi drivers protest Uber at Nathan Phillips Square, convinced the medallion would hold its value. It did not. But the drivers who learned the new platform are still driving. The ones who fought the technology are not. Law firms are businesses, and AI improves margins. No managing partner will apologize for that, and none should have to. Whether AI eventually reshapes the economics of the leveraged firm itself is still an open question. But law firms are also the only institution in this country that trains lawyers. There is no residency program, no alternative pipeline. The firm is the teaching hospital. I know this sounds idealistic. I also know that every senior lawyer reading this was trained by someone who believed the same thing. The elevator signs will come down. What replaces them cannot be the existing articling playbook with AI tacked on. It must be training that makes AI fluency a core competency from day one while ensuring that junior lawyers still do the unassisted analytical work through which judgment is built. I am writing this now because by the time my generation has the seniority to design that training, the apprenticeship that would have prepared us to design it well may no longer exist. ON MY first morning as a summer intern at a Toronto Bay Street firm, I stepped into the elevator at Bay Adelaide Centre and saw a laminated sign below the floor buttons: "The use of artificial intelligence tools for client matters is strictly prohibited." This message was on every floor, in every elevator bank. And the prohibition was right: the Law Society of Ontario's April 2024 white paper confirmed that duties of competence and confidentiality apply in full to AI. But Abraham Lincoln's law practice was limited by the number of books he could fit on his horse cart. Every transformation in legal information, from the printing press to Westlaw, has remade the profession. As University of Toronto law professor Benjamin Alarie has argued, AI is the next such transformation, and every law student should be learning to use it. The elevator signs will come down. The harder question is whether firms can train lawyers to use AI without hollowing out the apprenticeship that teaches them to think without it. I am graduating this spring. So are 4,000 other students from Canada's 24 accredited law schools. Some of us financed three years of tuition through OSAP. Some through family. Some worked jobs that had nothing to do with the law. We are not asking whether AI will take our jobs. Rather, we are asking whether the apprenticeship will make us competent enough to keep them. Every lawyer I admire says the same thing: legal training does not end with law school; it begins there. The classroom teaches doctrine, but the practice teaches judgment. You learn to advise a frightened client by watching a senior lawyer do it first and then doing it yourself, badly, and then doing it again. Every serious legal system institutionalizes this apprenticeship. Canada calls it articling; Germany, the Referendariat; and England requires two years of qualifying work experience. Competence is built through supervised repetition. The question is: what happens when those repetitions disappear? A&O Shearman deployed Harvey AI across its 3,500 lawyers. Thomson Reuters reported that AI adoption among legal organizations nearly doubled in a single year. Goldman Sachs estimated that 44 percent of legal tasks could be automated, the highest exposure among the professions surveyed. Legal judgment is built through repetition of exactly the tasks large language models do best: due diligence memos, contract review, and case-law summaries. Reviewing a contract AI has already flagged is a fundamentally different cognitive act from reviewing one cold