Law Schools Must Move Faster on Teaching AI in Legal Practice

I have long maintained that the refrain “AI won’t replace lawyers” is wrong.

To say artificial intelligence won’t replace lawyers, we would need to know AI’s ceiling. We don’t. The evidence instead points the other way: better foundational models, stronger legal tools, automation of entry-level work, and agentic systems capable of executing increasingly complex workflows.

The legal profession is changing quickly, and it may change in fundamental ways we can’t yet predict. In that environment, comforting slogans aren’t enough. If law schools fail to train students for the tools, workflows, risks, and market pressures already reshaping legal practice, they aren’t preserving traditional legal education. They are neglecting one of its core obligations.

Over the past year, I have taught artificial intelligence and law to more than 1,000 law students nationwide and globally, including through the nation’s first required law school AI course at Case Western Reserve University School of Law, the nation’s second at Mississippi College of Law,the first legal AI education program in East Africa, and programs at Washington University, the University of Washington, and Southwestern Law School.

During this time, I have learned several lessons: Law schools are moving too slowly, many administrators still don’t fully understand what meaningful AI education requires, and legal education hasn’t yet caught up to legal practice.

Three Distinct Pillars

Comprehensive AI education should generally include three pillars: AI in legal practice, AI ethics, and AI regulation. Law schools too often reduce AI education to abstract ethics warnings, policy discussions, or a standalone AI regulation course while overlooking the more practical need to teach students how to use AI tools responsibly and effectively.

AI in legal practice. This pillar cuts across practice areas. It is increasingly comparable to teaching students how to use traditional legal research platforms such as Westlaw and LexisNexis. No law school would suggest that students can graduate without meaningful exposure to those tools.

The same logic now applies to generative AI tools, including Harvey, Legora, Spellbook, Relativity, ChatGPT, and Claude. Law schools don’t dictate which tools lawyers use in practice. But they should monitor how legal technology is developing and prepare students for the tools that are already part of everyday legal work.

That preparation should be hands-on. Students need to learn how these tools create value—and how they fail. Students should encounter hallucinations, citation errors, and overconfident outputs before they confront those same issues in practice.

AI ethics and limitations. Students need to understand their professional obligations; it’s not enough for instructors simply to tell students not to over-rely on ChatGPT or not to input sensitive information into a publicly available chatbot. Those are important warnings, but they don’t capture the full complexity of potential risks to using these tools.

In midsized and large firms, for example, many lawyers may contribute to a document before it is filed, and every signatory may face Rule 11 sanctions exposure. We have already seen sanctions cases in which one lawyer inserted fake AI citations, but others who signed the filing were also sanctioned. Students should understand that filing responsibility in modern practice is shared.

The problem is also becoming more sophisticated. Hallucinations are no longer limited to nonexistent case citations. A source may be real, but the proposition, quote, or parenthetical attributed to it could still be wrong.

AI regulation. This pillar, which includes state and federal regulation and judicial governance, can no longer be treated as a niche elective. AI-related regulation and litigation now cut across practice areas as the technology becomes embedded in business, government, and society. Lawyers will increasingly advise clients on AI procurement, deployment, risk allocation, privacy, discrimination, intellectual property, employment, consumer protection, discovery, and professional responsibility. Even students who never practice “AI law” will encounter AI-related legal issues in ordinary matters.

Moving Slowly

Though AI may be one of the fastest-moving technologies in modern history, law schools often rely on lengthy approval processes and committee structures to introduce new courses and programs. That model is poorly suited to a technology changing this quickly.

Keeping up with AI model updates, legal tech products, ethics guidance, and regulation is effectively a full-time job. AI curriculum content can become outdated month-to-month, and students often know more about current AI products and capabilities than legal institutions do. Students are leaving universities having used these tools regularly—and law schools should be ahead of the curve.

Necessary Expertise

There was a moment in 2024 when some suggested that lawyers would need to learn to code—that claim was overstated. Law students don’t need Python or another programming language to use AI effectively. They need to understand legal workflows, where AI tools fit, and where they do or don’t add value.

Law schools also can expose students to “vibe coding,” or the use of natural language prompts to build simple legal technology applications. The point isn’t to turn students into software engineers but to help them identify legal workflow inefficiencies and think creatively about technology.

This is also the case with “prompt engineering,” meaning the practice of crafting instructions to guide an AI system toward a useful output. Its importance is often overstated today. Basic prompting can be taught quickly, and both legal tech products and general-purpose AI models are lowering the threshold for effective use. Judgment is the more important skill: understanding the task, choosing the right tool, framing the problem clearly, and verifying the result.

Abandon Old Claims

If the technology becomes capable enough, legal tech companies will have strong incentives to capture more of the legal services value chain rather than merely sell assistive software.

Few companies will say that openly, because it would slow adoption and alarm buyers. But it would be naive to assume the long-term ambition stops there.

Law schools should prepare students for that uncertainty, including asking the hard questions about access to justice, software-as-a-service-based legal service delivery, and the future structure of legal practice. As AI agents become more integrated into legal platforms, lawyers may spend less time performing discrete legal tasks from scratch and more time working in and alongside AI-enabled systems.

Legal work may increasingly be delivered through platforms, automated workflows, and hybrid lawyer-AI models. In that environment, lawyers may be asked to review AI-generated work, supervise automated processes, design legal workflows, and evaluate platform outputs. That makes training on and alongside AI platforms essential.

That is also why law schools should stop using the phrase “AI won’t replace lawyers.” It’s more honest to say AI is already changing legal work in unprecedented ways, and that no one can responsibly say with confidence where that change ends.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.

Author Information

Oliver Roberts is CEO & founder of Wickard.ai, co-director of the WashU Law AI Collaborative, an adjunct Professor of Law at Washington University in St. Louis School of Law, and managing attorney at The Roberts Legal Firm.

Interested in writing? Review our author guidelines, and submit pitches to Insights@bloombergindustry.com.