Artificial intelligence is improving both quickly and in ways that increasingly matter for legal practice and legal education. One of the most important developments has been the rise of reasoning models to supplement large language models (LLMs). Earlier, LLMs were best understood as systems for fast, fluent text generation. But newer reasoning models are designed to spend more time on planning, deduction, and problem-solving. Models now also include agents that can use tools to look up information on the Internet, reducing the likelihood of hallucinations.

　　The result is an AI system that can deliberate for far longer than earlier models, draft extended memos, and solve increasingly complex problems. Just last week, Timothy Gowers, a Fields medalist, said that ChatGPT 5.5 had solved an open math problem important enough to be part of a graduate student’s doctoral thesis. Some now predict that AI programs will soon be able to work through problems for days or even weeks, producing correspondingly more impressive results.

　　These improvements profoundly affect the creation of legal information. In some contexts, frontier models can now answer online student questions with speed, fluency, and accuracy that rivals what a professor can provide on the spot. They can also produce a series of legal memos in a day that would have required a research assistant days, if not weeks, of effort.

　　What does this mean for legal education? The first question is how it will change the legal profession, because legal education is downstream of the professions needs. Clients will not pay lawyers to do tasks that machines can perform more cheaply.

　　The rise of analytic legal machines will not kill all the lawyers, but it will transform what lawyers do, what clients will pay for, and what law schools must teach. Lawyers must continue to adapt their work as AI capabilities evolve. Economists, like Alex Imas and Luis Garicano, think about jobs in terms of supply and demand, and their analysis can be applied to legal work as well.

　　Start with the demand side. Demand will shift away from lawyers whose principal value is routine analysis because machines will increasingly perform that work better, faster, and more cheaply. But clients buy, not just legal words, but also the analysis that AI can readily produce. They buy judgment, reassurance, and confidentiality. Currently, at least, they need someone who can argue in court and sign an opinion.

　　Thus, it is true that items such as the first draft of a memo or a standard contract clause will become increasingly cheap to produce. But the client meeting after the memo will still take place. Most, if not all, clients will still want the final decision about whether to sue or take reputational risk in each situation to be made with human input. The trust required to make those decisions on behalf of or in concert with a client is not cheap.

　　Machines can already isolate all the judges opinions and thus frame an argument designed to move the relevant decision-maker. And soon, machines will be able to perform very sophisticated analysis using decision trees and probability assignments for different scenarios. Nevertheless, lawyers who can bond with other human beings and help them make challenging decisions in the face of uncertainty will become more, not less, valuable. Those are the kind of lawyers that law schools will need to produce.

　　Law and economics also suggest that AI will not merely reduce demand for lawyers. Lowering the cost of routine legal analysis will likely expand the market for legal services among individuals and businesses that now forgo them. This development will be a great boon to democratizing legal planning and, thus, liberty. But cheaper legal output may also still have a quality problem. Clients, particularly unsophisticated ones,often cannot tell whether a legal answer is sound for their particular circumstances. Lawyers will therefore become more valuable as certifiers of legal quality.

　　Now turn to the supply side. Economists recognize that jobs are not isolated tasks, but bundles. Some bundles are weak. If you can peel off a task and automate it, you can leave the human with less to do. Other bundles are strong because the tasks belong together. Separating them creates too much coordination cost, too much risk, or too much loss of accountability.

　　Legal practice contains many strong bundles. AI can prepare outlines and analyze transcripts, but courtroom advocacy still requires credibility, emotional intelligence, and the ability to adjust when a judge or opposing counsel does something unexpected. William Butler Yeats famously asked, “Who can separate the dancer from the dance?” suggesting that there are activities where the creation cannot be separated from the activity. So long as law demands generate situations that require real-time creative responses from humans, in-the-moment legal creativity will be required.

　　We need more training in oral and rhetorical skills. In practice, the future lawyer’s value will often lie in being able to orally defend and take responsibility for work that AI helped produce.

　　Law rests on judgment and responsibility, which also creates a bundle. A legal system cannot now run on anonymous machine output. Someone must sign the filing. Someone must advise the board. Someone must tell the client, “This is lawful,” “This is too risky.” And he must understand the output of machines and the law well enough to make that judgment.

　　Thus, the current hiring strength in BigLaw is not evidence that AI will do nothing—it is yet more evidence that law is still a strong bundle. Clients still need judgment, reassurance, confidentiality, advocacy, and someone who can take responsibility. But that does not mean every task inside the bundle will retain its old value. Currently, AI still preserves many roles for lawyers. But it is already changing what the lawyer is paid to do. And, of course, current results do not guarantee future security. AI is improving monthly, and with that improvement, it will colonize ever more tasks.

　　Law schools should respond in several ways. First, they must ensure that AI literacy becomes part of professional formation. Even as AI replaces coding, companies continue to hire software engineers who can harness AI to write code. Being a master of AI is no less valuable for lawyers. Students need to learn how to use AI, but also how not to be used by it. With this partner, they must learn verification, confidentiality, and the difference between a plausible answer and a responsible judgment. Certification of the quality of machine work will remain essential. That means they must still be fully trained and independent as lawyers, but also able to use and evaluate AI constantly.

　　Thus, some parts of the law school curriculum will continue to focus on providing the independent legal skills needed to judge AI outputs. Only then can lawyers take responsibility and play the role of certification, which will become even more essential. I imagine, for instance, that the content of the doctrinal courses in the first year may be unchanged even if AIs become useful tutors to help students learn legal rules.

　　Second, assessment must change. If students practice in an AI-rich world, we should stop designing assessments for every kind of course as though AI does not exist. Some assessments in some courses should require students to use AI. They must then evaluate, supplement, and stand behind their output. Seminars, for instance, should include extended oral presentations so that the professor can better assess the students capacity to defend machine-assisted work fluently, as well as their classmates’ capacity to probe the product.

　　Third, law schools should teach the strong bundle. That means more exercises that bring together doctrine, facts, ethics, strategy, and institutional roles.A student should be trained not only to learn the summary judgment doctrine, but also to apply it. He or she should have to decide whether filing the motion advances the client’s real goals.

　　Another reason for doing this is that law firms may no longer do as much of this kind of training as before. AI endangers the apprenticeship period by automating the tasks through which junior associates historically acquired judgment. In the old model, training was bundled with productive work. But if AI performs those tasks faster and more cheaply, firms may reduce the involvement of these associates, leaving training as a separate investment that law schools may need to provide.

　　Fourth, we need more training in oral and rhetorical skills. In practice, the future lawyer’s value will often lie in being able to orally defend and take responsibility for work that AI helped produce. A focus on oral skills will be a sharp reversal of how the law school class has evolved in the past decades. The Socratic method, which involves cold calling students, has declined in popularity. Education has become more consumer-oriented, and most students dislike public grilling because it may embarrass them in front of their peers. Moreover, law professors today are less connected to the bar, where the give-and-take of oral argument is the lawyers’ daily bread, and more integrated into the Academy, where it’s quite normal to stand at the front of a classroom giving lectures from PowerPoint. But in a world where AI machines will be whirring behind the scenes and providing much legal analysis, talking persuasively and engagingly will be an ever more essential part of a lawyer’s job.

　　Finally, AI will move us toward a more relational economy because humans still value human relations. Client counseling, negotiation, interviewing, leadership, and ethical judgment are key skills in this relational world. They are the parts of the legal bundle least likely to be commoditized.

　　Of course, if AI continues to improve rapidly, as I expect, more radical reforms may be required. Some legal work today is protected less by technical necessity than by legitimacy, convention, and the human need for responsibility. It is still hard to imagine an AI arguing in court, counseling a board, or standing before a client and taking responsibility for a judgment. But conventions can change when technology changes what seems possible, useful, and legitimate. One can imagine, for instance, that in arbitration, the parties may choose to restrict themselves to AI lawyers and an AI judge to enhance predictability and thus the value of the contract. But the law school that begins now to train lawyers in judgment, persuasion, and the intelligent use of machines will be far better prepared for the next round of changes demanded by the accelerating progress of AI.