What is the irreducibly human element in legal education when AI can pass the bar exam, generate effective lectures, and provide personalized learning and academic support? This Article confronts that question head-on by documenting the planning and design of a comprehensive transformation of a required doctrinal law school course—first-year Contracts— with AI fully embedded throughout the course design. Instead of adding AI exercises to conventional pedagogy or creating a stand-alone AI course, this approach reimagines legal education for the AI era by integrating AI as a learning enhancer rather than a threat to be managed. The transformation serves Mitchell Hamline School of Law’s access-driven mission: AI helps create equity for diverse learners, prepares practice-ready professionals for legal practice transformed by AI, and shifts the institutional narrative from policing technology use to leveraging it pedagogically.
This Article details the roadmap I have followed for AI integration in a course that I am teaching in Spring 2026. It documents the beginning of my experience with throwing out the traditional legal education playbook and rethinking how I approach teaching using AI pedagogy within a profession in flux. Part I establishes the pedagogical rationale grounded in learning science and institutional mission. Part II describes the implementation strategy, including partnerships with instructional designers, faculty innovators, and legal technology companies. Part III details a course-wide series of specific exercises that develop AI literacy alongside doctrinal and skill mastery. Part IV addresses legitimate objections about bar preparation, analytical skills, academic integrity, and scalability beyond transactional courses. The Article concludes with a commitment to transparent empirical research through a pilot study launching in Spring 2026, acknowledging both the promise and the uncertainty of this pedagogical innovation. For legal educators grappling with AI’s rapid transformation of both education and practice, this Article offers a mission-driven, evidence-informed, yet still preliminary template for intentional change—and an invitation to experiment, adapt, and share results.
INTRODUCTION
I stop mid-sentence.
It’s a typical Tuesday evening in my Secured Transactions course—Article 9 of the Uniform Commercial Code (Code). I’ve taught the course dozens of times since I started teaching law in 2002. Not much has changed in how I teach the course, except that the class moved online when William Mitchell College of Law (WMCL) launched its hybrid JD program in 2015, and I am now teaching from my basement office. And while the structure and pedagogy of the class are largely the same, I have made small updates to the content to keep pace with four sets of amendments to Article 9 (2010, 2018, 2020, 2022). I’m lecturing about the third topic in the course, which is perfection—how a secured creditor achieves a public claim in collateral. I answer a few student questions, getting ready to work through a set of problems in which the students apply the Code to advise hypothetical clients.
Then I ask my students a question that I had been wondering about a lot: Is there any need for me to teach them Secured Transactions in the world of generative artificial intelligence (AI)?
I comment that AI can generate a lecture,1 a bot could answer students’ questions and give them hypotheticals and feedback on their responses, and AI could administer and grade assessments and assign final grades to comply with the school’s mandatory mean.2 Perhaps my twenty-four year teaching career should come to an end, and Professor AI could teach this and other classes.3
* Gregory M. DuhlProfessor of Law, Mitchell Hamline School of Law (MHSL). I thank MHSL instructional designer Amanda Soderlind for her patience (sorting through my frequent emails with new ideas), guidance (figuring out how to navigate this new pedagogy), and ongoing reminders that we cannot sacrifice good pedagogy in the name of innovation. I also thank my colleague Professor Anthony Niedwiecki, whose Business Organizations students brought their contagious enthusiasm for his AI tools to Secured Transactions every Tuesday evening, and who has joined me on this pedagogical AI journey. I am appreciative of MHSL law student Hallie Wiederholt, who voluntarily engaged with this article and shared insights arising from psychological research and her law school experience; she is representative of the many MHSL students who read a draft and shared their thoughts about the course redesign. Last, I extend my deepest gratitude to my MHSL co-collaborators in designing and launching the hybrid JD program (Annie Gemmell, Jim Hilbert, Eric Janus, Peter Knapp, Mehmet Konar-Steenberg, and Kelly Von Ruden) who, over a decade later, have encouraged me to do this work and helped me recapture my innovative spirit.
1 See, e.g., Ronald R. Danault, Adapting to AI: The Evolving Role of Faculty in Higher Education, U.N.H. SCHOLARS REP. 6 (2025), https://scholars.unh.edu/faculty_pubs/2156/.
2 See Sean A. Harrington, Introducing QuizBot an Innovative AI-Assisted Assessment in Legal Education (Dec. 10, 2024) (unpublished manuscript), https://ssrn.com/abstract=4975804 (on the idea that a bot can give hypotheticals and ask questions and provide feedback, answer questions, and administer/grade assessments); Zhihui Zhang et al., The Role of Generative AI and Hybrid Feedback in Improving L2 Writing Skills: A Comparative Study, in INNOVATION IN LANGUAGE LEARNING & TEACHING 1, 2 (2025) (supporting the idea that AI provides rapid “nuanced, goal-oriented, and human-like feedback,” which would allow a bot to handle at least a significant part of the assessment process); Kevin L. Cope et al., Grading Machines: Can AI Exam-Grading Replace Law Professors? 23 (Va. Pub. L. & Legal Theory Rsch. Paper No. 2025-80, Dec. 4, 2025), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5851362 (finding AI extremely consistent with human professor’s grades, showing high correlation of up to .93 when the AI is given a detailed grading rubric).
3 See Nick Ladany, Meet the AI Professor: Coming to a Higher Education Campus Near You, FORBES (Oct. 3, 2025, 16:02 EST), https://www.forbes.com/sites/nicholasladany/2025/10/03/meet-the-ai-professor-coming-to-a-higher-education-campus-near-you/ (discussing the capabilities of “Professor AI”).
The students protest and categorically dismiss my suggestion. They insist that when I teach the Secured Transactions material, I do so with clarity, subtlety, humor, and compassion. AI cannot—well, at least not yet. But were they merely stroking my ego? Most students do not even attend these live office hours because they can watch the recordings at their convenience. Am I any better a teacher than a bot?
After class that night, I contemplate how I am doing my students a disservice by not keeping pace with AI’s transformation of higher education and the legal profession. I write to Amanda Soderlind, my instructional designer at Mitchell Hamline School of Law (MHSL), with whom I have partnered for seven years in designing online and hybrid courses. I share the same reflections with her that I shared with my class. The next day, she fed my lecture notes from the previous evening into an AI tool, and it produced a video of my lecture. There were a couple of errors, but within about thirty minutes of critique and improvement, she generated an accurate and engaging presentation of the previous evening’s lecture, complete with visuals of the rules and hypotheticals. In at least certain ways, it was more effective than what I had presented because I have long struggled with visual-spatial processing and therefore create few visuals when I teach, though I know that visual examples appeal to spatial learners.
Research on AI-generated educational videos supports both the promise and the limitations of this approach. In a controlled experiment with 447 participants, researchers found that while students prefer human-made teaching videos in terms of learning experience, they achieved equally high learning outcomes when watching AI-generated videos.4 This finding captures the tension at the heart of this article: AI can produce effective learning outcomes, but human facilitation and connection remain essential to the educational experience. It’s clear to me, though, that faculty roles must evolve to leverage AI’s capabilities while preserving the irreplaceable human elements of teaching.
Here is the thing about Secured Transactions: most students take it not because they are fascinated by Article 9, but because it is tested on the bar exam. That is unfortunate; but, if I am lucky, I usually convince at least a few students by the end of the course how fascinating and practical Secured Transactions is (e.g., if they default on a car loan, how do they prevent the bank from repossessing the car?).5 But for the sole purpose of bar preparation, Amanda had just demonstrated that AI could replace me. A student could watch AI-generated lectures, practice hypothetical essay questions with a bot, and walk into the bar exam adequately prepared. The students did not need me.
This realization confounded what I had already realized the previous year: AI itself could generate a passing bar-exam essay.
4 See Torbjørn Netland et al., Comparing Human-Made and AI-Generated Teaching Videos: An Experimental Study on Learning Effects, 224 COMPUTERS & EDUC. 1,10 (2025).
5 Starting in July 2026, the National Conference of Bar Examiners is no longer including Secured Transactions as one of the topics tested on the Multistate Essay Examination or the NextGen Bar Examination. Press Release, Some Subjects to Be Removed from MEE in 2026, NAT’L CONF. BAR EXAM’RS (July 17, 2023), https://www.ncbex.org/news-resources/some-subjects-beremoved-mee-2026. I might be out of a job unless I can convince students to take Secured Transactions because of how fascinating and practical the course is.
I know this because I tested it informally in my own classroom in Spring 2024. I told my small Secured Transactions class that they should select one student to produce an answer to the last assigned bar-exam-style essay using solely AI, but not to tell me whom they selected until I finished grading the essays. I signed out of Zoom and let them designate the AI user. After I graded the essays, the AI user came forward. Her AI-generated essay had received full credit, five out of five. The machine had performed as well as any human student. Even if AI produced an essay that scored 3.5 or four out of five, any student who relied solely on AI in writing the four course essays was, at least according to the Canvas gradebook, performing well in the class.
The implications of this experiment had already kept me awake at night. How do I effectively teach students who would perform better if they turned in AI-generated work than student generated work? Of course, I was not naive enough to ignore that students were already turning in AI-generated and AI-assisted work. But now these nighttime ruminations were compounded by my realization that AI might be as effective a teacher as I am. Consider the logical endpoint: Research has demonstrated that AI can pass the bar exam.6 The bar exam is designed to test minimum competence to practice law.7 If AI is minimally competent to practice law, and AI can also teach the material that prepares students for the bar exam, then we have Faculty AI teaching Student AI. Where, exactly, do humans fit into this picture?
This concern extends far beyond questions about student cheating, technology use in the classroom (either to “catch” student AI use or to facilitate it), or effective pedagogy. It is a concern that at its core is about the professor–student relationship, how that relationship takes shape in legal education, and how it connects to the legal profession. If AI can perform the role of professor and student, at least at a level sufficient to meet our profession’s minimum standards, then we must ask what is the irreducibly human element in legal education and legal practice?
6 See Daniel Martin Katz et al., GPT-4 Passes the Bar Exam, 382 PHIL. TRANSACTIONS ROYAL SOC’Y A 1, 11 (2024), https://doi.org/10.1098/rsta.2023.0254 (In 2023, researchers demonstrated that GPT-4 could pass the Uniform Bar Exam, scoring approximately 297 points, which is well above the passing threshold in all jurisdictions, and also outperformed the average human testtaker.); Pablo Arredondo, GPT-4 Passes the Bar Exam: What that Means for Artificial Intelligence Tools in the Legal Profession, STAN.L.SCH.: LEGAL AGGREGATE (Apr. 19, 2023), https://law.stanford.edu/2023/04/19/gpt-4-passes-the-bar-exam-what-that-means-for-artificialintelligence-tools-in-the-legal-industry/ (“We found that while GPT-3.5 failed the bar, scoring roughly in the bottom 10th percentile, GPT-4 not only passed but approached 90th percentile. These gains are driven by the scale of the underlying models more than any fine-tuning for law. That is, our experience has been that GPT-4 outperforms smaller models that have been fine tuned on law. It is also critical from a security standpoint that the general model doesn’t retain, much less learn from, the activity and information of attorneys.”); See also Eric Martínez, Reevaluating GPT-4’s Bar Exam Performance,33 ARTIFICIAL INTELLIGENCE &L. (2024), https://doi.org/10.1007/s10506-024-09396-9 (Of particular note is the finding that initial percentile estimates of passage rate were likely overinflated, particularly on essay components, though confirming GPT-4 still achieved passing competency.).
7 Danette Waller McKinley & Beth E. Donahue, The Testing Column: Measuring Competence: Assessment of Knowledge and Skills on the Bar Exam, 93 THE BAR EXAM’R 21, 21 (Fall 2023), https://thebarexaminer.ncbex.org/article/fall-2023/the-testing-column-fall23/ (stating that the Uniform Bar Examination “is used in 41 jurisdictions to determine whether those seeking to enter the practice of law possess the minimum level of competence to serve the public”).
I am not alone in asking this question. The AI transformation is accelerating across the profession8: while firm-wide AI implementation stands at 21 to 24 percent overall, individual attorney usage of AI tools has surged to 55 to 81 percent, with in-house counsel leading adoption.9 Large law firms are driving this shift—among firms with 500 or more attorneys, adoption rates range from 48 to 100 percent depending on survey methodology, compared to just 20 to 32 percent among small firms.10
Law schools are scrambling to respond: 55 percent now offer AI-specific courses, and 62 percent have integrated AI into first-year curricula.11 However, in building AI-specific courses and adding AI modules to existing courses, legal education is not taking advantage of the opportunity to fundamentally reimagine pedagogy.
8 See Arrendondo, supra note 6 (“The rate of progress in this area is remarkable. Every day I see or hear about a new version or application. One of the most exciting areas is something called Agentic AI, where the LLMs (large language models) are set up so that they can ‘themselves’ strategize about how to carry out a task, and then execute on that strategy, evaluating things along the way.”).
9 Midhat Tilaway, AI in Law Statistics 2025: 55% of Lawyers Already Use AI and Adoption Is Accelerating, ALLABOUTAI (Oct. 23, 2025), https://www.allaboutai.com/resources/aistatistics/ai-in-law/ (AI adoption rates vary significantly depending on measurement methodology. Firm-wide implementation policies stood at 21 percent of law firms in 2025, though this represented a slight decline from 24 percent in 2023, suggesting that initial enthusiasm has given way to more deliberate, strategic adoption. However, individual attorney usage tells a different story: 55 percent of attorneys in law firms now use AI tools for legal work, and 81 percent of in-house counsel have adopted AI tools. Among attorneys who use AI, 45 percent use it daily and 40 percent use it weekly.).
10 Among firms with 500 or more attorneys, adoption rates range from 47.8 percent to 100 percent, depending on survey methodology. See Mark Calaguas, 2024 Artificial Intelligence TechReport, AM. B. ASSOCIATION (Apr. 25, 2025), https://www.americanbar.org/groups/law_practice/resources/tech-report/2024/2024-artificialintelligence-techreport/ (reporting 47.8 percent AI usage among firms with 500+ employees); Soledad Atienza, AI Transformation in the Legal Sector Begins in Law Schools, PHYS ORG (Mar. 25, 2025), https://phys.org/news/2025-03-ai-legal-sector-law-schools.html (finding 100 percent of firms with 500+ lawyers reported AI integration compared to 68 percent of small firms not yet using AI in a survey of 333 firms). By contrast, in another survey, small firms with 1 to 100 lawyers show only 20 to 32 percent adoption rates. See Tilaway, supra note 9. The broader measurement of “legal professionals”—which includes in-house counsel who adopt AI at much higher rates—shows adoption jumping from 19 percent in 2023 to 79 percent in 2024. See CLIO, 2024 LEGAL TRENDS REPORT 26–27 (2024).
11 ABA TASK FORCE ON LAW AND ARTIFICIAL INTELLIGENCE, AI AND LEGAL EDUCATION SURVEY RESULTS: 2024 1, 11 (2024), https://www.americanbar.org/content/dam/aba/administrative/office_president/task-force-on-law-and-artificial-intelligence/2024-ai-legal-ed-survey.pdf (“Moreover, an overwhelming majority (83%) reported the availability of curricular opportunities, including clinics, where students can learn how to use AI tools effectively.”). The sample size was small; only twenty nine law schools responded to the survey. It is possible that, among those law schools that responded, there is more AI integration in the curriculum than at schools that did not.
Meanwhile, students are not waiting for their institutions to catch up. They are using generative AI to search for legal information, brief cases and create outlines, prompt chatbots to quiz them on doctrine, and—for those with access to premium tools—building custom study assistants tailored to their courses and to the bar exam.12 The transformation is happening at breakneck speed, whether or not legal educators choose to embrace it. We have a choice as faculty: either lead this transformation in legal education or simply be overtaken by it.
I’m almost 57, well into the back half of my teaching career. This AI transformation raises existential questions for me, about my purpose as an educator, the value I add, and my future in legal education. But more urgently, it raises questions about my present. If I can be replaced, or if I am already teaching in a way that AI could replicate, what does that say about how I have been teaching? More importantly, what does it say about how I should be teaching?
What is certain is that I have to evolve as an educator in a world where AI is infiltrating legal education and the practice of law. But I also have to figure out the why. When WMCL sought a variance from the American Bar Association for a hybrid JD program in 2013, we confronted the same question: Why? Even with a pedagogically sound proposal for a blended-learning JD program with a lot of appeal to prospective students, what was driving our request for a variance? The answer now is the same as the answer then. Our mission at WMCL (which got the variance) and the successor school MHSL have been to increase access to legal education and access to legal services.13
Our current vision, mission, and values:
- Vision: A legal system that is just and accessible to all.
- Mission: To provide a rigorous legal education through broad access and support for students’ holistic growth, and to build practice-ready professionals with passion for law and justice.
- Values: Courage and independence; Commitment and accountability; Inclusion and belonging; Community and collegiality.
We are an access school and generally not training students to go into big firms. Our students and graduates reflect this mission: they come from geographical areas without an accessible law school and where there are not sufficient lawyers; they are Native American and make up the largest Native American student
12 See, e.g., Austin Gergen, Navigating the AI Revolution in Law School: Lessons from the Front Lines, ABA J. (Apr. 30, 2025, 10:23 CDT), https://www.abajournal.com/voice/article/navigating-the-ai-revolution-in-law-school-lessons-from-the-front-lines.
13 See generally Eric S. Janus, The “Worst Idea Ever!” — Lessons from One Law School’s Embrace of Online Learning, 70 SYRACUSE L. REV. 13 (2020) (detailing the mission-driven design of the hybrid JD program).
population of any law school in the country;14 they become public defenders, county attorneys, and judges; they work in government agencies and nonprofit organizations; they launch solo practices serving clients who might otherwise go without representation. Each year, 15 to 20 percent of new graduates choose solo or small firm practice, and on average nearly one in five secure judicial clerkships.15 I wondered, just because some lawyers and law schools are using AI, does that mean our students need it?
Well, a colleague of mine, Eleanor Frisch, who is a disability rights scholar in her first year teaching at the law school, suggested in a casual comment that AI could help neurodiverse students and students with disabilities learn. I have also realized through additional research that AI could help first-generation students succeed in law school and level the playing field for students who come from underrepresented populations. AI could enable legal educators to accommodate different learning modalities—not as a retrofit or accommodation, but as embedded classroom design. AI could help us achieve better learning outcomes through personalized learning, student practice, and immediate feedback, and open up more opportunities for academic support and bar preparation. And critically, for students training to do social justice work, work in the courts, or public service, AI could become a tool to improve access to justice and legal services.
AI isn’t just a shiny new technology. It’s a pathway to access. And it happens to align with our institutional mission.
My Perspective as a Neurodiverse Professor
But there was something else. I am neurodivergent.16
Standing in front of a classroom or even being in a classroom is exhausting for me. Office hours drain me. I have taught for twenty-four years, and I’ve been told I’m good at it. Student evaluations consistently note that I’m engaging, knowledgeable, intelligent, funny, innovative, passionate, approachable. But what students see—the performance—requires constant effort to mask, to maintain, to sustain.17
14 Press Release, Mitchell Hamline Sch. of L., Recruitment Efforts Put Mitchell Hamline First in the Nation in Native Enrollment (Nov. 25, 2024), https://mitchellhamline.edu/news/2024/11/25/recruitment-efforts-put-mitchell-hamline-first-in-the-nation-in-native-enrollment/.
15 See Solo and Small Firm Practice, MITCHELL HAMLINE SCH. LAW, https://mitchellhamline.edu/careers/solo-and-small-firm-practice-program/ (last visited Nov 26, 2025); see also ABA Employment Summary for 2024 Graduates, MITCHELL HAMLINE SCH. LAW, https://mitchellhamline.edu/careers/wp-content/uploads/sites/10/2025/04/ABA-employment-summary-2024.pdf (last visited Dec. 12, 2025).
16 Cf. Gregory M. Duhl, Over the Borderline—A Review of Margaret Price’s Mad at School: Rhetorics of Mental Disability and Academic Life, 44 LOY. U. CHI. LJ. 771 (2013).
17 See GREGORY M.DUHL, STRENGTHS AND CHALLENGES OF AUTISTIC FACULTY IN HIGHER EDUCATION A WHITE PAPER 6–8 (Oct. 2025) (outlining the experiences of neurodivergent faculty in higher education, including their valuable cognitive contributions and the systemic barriers they face, particularly the exhausting cognitive labor of masking and the difficulty in accessing institutional accommodations).
I am energized when I’m behind a computer, figuring out how to facilitate learning. I come alive when I’m inventing pedagogy, building curricular frameworks, and designing course assignments. I am motivated by the creative work of teaching, the architecture of learning. The performative work of teaching—the standing, the speaking, the immediate responsiveness required in synchronous classes—drains me.
For years, I accepted this as simply part of what otherwise is a wonderful and fulfilling career. If I wanted to be a professor, I had, to a lesser or greater extent, to perform. There was no alternative. Now, AI offers an alternative.
For example, if I could design a class where students engage in individual Socratic dialogues rather than performing in front of their peers, I reduce my own need to mask while simultaneously creating a learning environment where neurodiverse students don’t have to mask either. More generally, if I could front-load the intellectual work of course design—building the AI tools, crafting the exercises, structuring the learning experiences—I could spend less energy on performance and more energy on what I do best: understanding individual students’ successes and struggles, intervening where appropriate, and facilitating their growth.
I have long understood that neurodivergent professors bring diverse perspectives, enhance problem-solving, and foster innovation through their unique ways of thinking.18 We offer deep focus and pattern recognition, which can translate into innovative teaching methods. We can provide crucial empathy and mentorship for neurodivergent students. We challenge stigma through our visibility. We advocate for more inclusive classrooms.19
But we can only do this if we are not exhausted by the expectation that we comply with neurotypical professorial norms. AI has the potential to both serve neurodivergent students and empower neurodivergent professors. It allows us to teach from our strengths rather than compensate for our weaknesses. As research shows, educators who are neurodivergent build barriers they face, particularly the exhausting cognitive labor of masking and the difficulty in accessing institutional accommodations). 20 AI offers tools to reduce those challenges.
18 See id.
19 See Katherine Silver Kelly, Be Curious, Not Judgmental: Neurodiversity in Legal Education, 78 ARK. L. REV. 245, 276 (2025) [hereinafter Be Curious]; see also Joseph Park, AI Through the Lens of Neurodiversity, MEDIUM (Nov. 23, 2023), https://medium.com/digital-architecture-lab/ai-through-the-lens-of-neurodiversity-3134c7ec11a7 (“Clearly, there’s value in crafting AI that thinks more like humans. However, this doesn’t imply the resulting ‘intelligence’ should be homogenous. Humans have their unique strengths, while AI can continue to exceed human capabilities in areas like pattern recognition. Rather than perceiving this as a threat, neurodivergence offers insights. By expanding our comprehension of human intelligence’s diverse forms, we can embrace AI as another legitimate form of intelligence and appreciate its distinct contributions, which will lead us to a more informed and constructive dialogue about our shared future.”).
I thought about how I could design an entire class, put in the majority of the work up front, and then make adjustments and give feedback as the class moved along. Student-centered learning in the truest sense. Less time performing, more time connecting with individual students about their challenges and successes. I was excited.
The Decision to Start Over
I decided to start my pedagogical journey with first-year students. Contracts is a second semester, first-year course at MHSL, a course I’ve taught yearly for most of the last two decades. It’s a foundational course, required for all JD students, and is heavily tested on the bar exam.
If I was going to do this, I was not going to dabble. I was not going to add an AI exercise or two to my existing course design and call it innovation. I was going to discard everything, including my casebook, my lecture notes, my exercises, my assessments. Twenty years of accumulated materials, refined over time, optimized through experience.
Gone.
I wanted to design the class as if I were teaching Contracts for the first time in a world where AI already exists and is pervasive in legal practice. I started my course design by establishing fundamental learning outcomes:
By the end of the course, students are able to:
- Articulate legal principles and holdings from contracts cases
- Identify common-law contract doctrine on choice of law, contract formation, defenses, interpretation, performance, and remedies
- Apply common-law contract doctrines to client hypotheticals
- Evaluate business contracts for potential risks and drafting ambiguities
- Draft a two-party employment agreement
- Represent a hypothetical client in resolving a commercial lease dispute
These objectives span the full range of Bloom’s taxonomy,21 from remembering and understanding doctrine to applying, analyzing, evaluating, and creating. They are the near identical outcomes
20 See Elizabeth Miller, Neurodivergent Educators Build Connections but Face Unique Challenges in Northwest Schools, OR. PUB. BROAD. (Dec. 16, 2023, 7:00 A.M.), https://www.opb.org/article/2023/12/16/neurodivergent-autism-adhd-teachers-educators-education-schools-disability-oregon/ (discussing the challenges of neurodivergent educators but their unique capacity of empathy for neurodivergent students).
21 See generally LORIN W. ANDERSON & DAVID R. KRATHWOHL, A TAXONOMY FOR LEARNING, TEACHING, AND ASSESSING: A REVISION OF BLOOM’S TAXONOMY OF EDUCATIONAL OBJECTIVES (2001) (revised version of foundational text on Bloom’s Taxonomy, which classifies cognitive processes hierarchically from lower-order skills like remembering, understanding, and applying, to higher-order skills like analyzing, evaluating, and creating).
to what I’ve always used for the first-year Contracts class. But I wanted students to achieve them using whatever AI tools I could find to improve my teaching and their learning and to give them a leg up with how legal practice is evolving.
I wrote an email to Amanda: “I am all in, and I need you to be as well.” It was a return to the playground.
When I first started teaching in 2002, everything was new. Every class was an experiment. Every assignment was a discovery. I had tools to explore, approaches to test, failures to learn from. It was joyful because it was playful—like a child with new toys, trying things just to see what happened.
Over time, teaching became more routine. I found approaches that worked and repeated them. The playfulness faded. Designing and teaching in our hybrid JD program (and then our blended learning program) brought back some of that joy. Teaching was playful again because I had new online tools to discover (e.g., breakout rooms, polling, chat), new challenges of distance to solve. But after thousands of Zoom classes and dozens of in-person, on-campus weeks with blended students, even that model had gotten somewhat stale over the past few years.
AI created an opportunity to be playful all over again. A new playground, new toys, a whole new world for which I could design a class.22 I am more energized. More animated. AI and solutioning how to integrate AI into my Contracts course has breathed new life into my teaching and has helped me access my “flow,” the optimal state for human productivity and engagement.
This is my “why”: To align AI integration with our institutional mission of access. To create more equitable learning environments for neurodiverse students, students with disabilities, first generation students, and underrepresented groups. To prepare students for the practice of law as it actually exists in 2026 and 2027 and will exist in 2030, not as it existed in 2025 or 2020. To reclaim the joy and playfulness of teaching.
And to answer the question that stopped me mid-sentence in Secured Transactions: What do I offer students that AI does not?23
22 Andrew Maynard, AI in Higher Education: Students Need Playgrounds, Not Playpens, FUTURE BEING HUMAN (Mar. 15, 2025), https://www.futureofbeinghuman.com/p/airplaygrounds-in-higher-education; Brian Piper, AI For U, Ep. 16: Imagining the Future of Higher Education with Andrew Maynard, ENROLLIFY (Mar. 13, 2025), https://www.enrollify.org/episodes/ep-16-imagining-the-future-of-higher-education-with-andrew-maynard (Maynard frames the debate of AI use in higher education as playground [AI as an enhancer] versus playpen [AI as cheating], which is apt and aligns with my perspective on the current dynamic in legal education.); see generally MIHALY CSIKSZENTMIHALYI, FLOW: THE PSYCHOLOGY OF POSITIVE EXPERIENCE (1990) (defining the optimal state for human productivity and engagement).
23 Maher Ghalayini, AI Isn’t Replacing Professors—It’s Redefining Them, MEDIUM (July 31, 2025), https://medium.com/@mghalayinisl/ai-isnt-replacing-professors-it-s redefining-them-7f7eeac72d4a (“Across campuses, faculty aren’t simply responding to AI. They’re reimagining how students learn, how research gets done, and what it means to teach in a digital-first world. From redesigned writing curricula to AI-assisted research, professors are becoming architects of a new academic era.”).
The answer isn’t that I deliver content better. AI can do that as well as I can, maybe better. The answer is that I can facilitate learning in ways AI cannot. I can design the experiences, curate the tools, monitor individual progress, intervene with targeted support, focus on higher-order cognition, model professional judgment, and teach ethical reasoning. I can build relationships with students based on trust and mutual growth, and uninhibited by their fear of being “caught” cheating under an outdated code of academic misconduct.
I don’t need to perform. I can be a guide or facilitator—or, more accurately, an architect of learning experiences and a mentor for individual growth. But to do that, I had to blow up everything I had been doing and start over. So I did.
Is this the “worst idea ever”?24 Maybe. I welcome the skepticism. But I joined a group of faculty and administrators about twelve years ago in coming up with what was previously the “worst idea ever” (WMCL’s hybrid JD program), and my track record with “worst” ideas suggests that this is worth a shot.
I do not yet know whether this approach will succeed. Conviction is not evidence. To find out, I am conducting a pilot research study alongside this course redesign, measuring student learning outcomes, self-efficacy, and experience with AI-integrated pedagogy. This Article describes the design and rationale; a subsequent article will report results. This Article offers hypotheses that I believe are well-grounded in pedagogical theory and institutional mission. But they remain hypotheses until we see the results.
Before getting into the details, let me clarify what this AI-integrated course looks like in practice. Students work with AI tools throughout the semester to enhance their learning—specifically, their writing, their research, their analysis, their drafting, and their negotiating. However, the final exam remains a traditional, closed-book, multiple-choice exam using actual Multistate Bar Exam questions, counting for one-third of their final grade. This design is intentional: the closed-book exam simulates the conditions students will face on the bar exam, where they must demonstrate their ability to recall and apply doctrine without technological assistance. Low stakes formative quizzes at the end of each unit prepare students for this exam. However, the bulk of the students’ grade comes from AI-enhanced learning activities throughout the semester. This approach avoids any breakdown in trust, as students know exactly when AI is permitted and when it is not, and the pedagogical reasoning for each context is transparent. And it does not try to pretend that AI has a place in every pedagogical context.
This Article proceeds in five parts. Part I articulates the pedagogical rationale for embedding AI into a required doctrinal course, advancing three core justifications: (a) creating access and equity through AI-enhanced pedagogy, (b) preparing practice-ready lawyers for an AI-transformed profession, and (c) shifting the dominant narrative from policing student AI use to designing a classroom that leverages AI for deeper learning. Part II describes the implementation strategy, detailing the four foundational resources that made this course design possible:
24 See Janus, supra note 13, at 14–15 (regarding the hybrid JD program as potentially being the “worst idea ever,” but actually succeeding in practice).
(a) partnership with in-house instructional design expertise, (b) learning from faculty innovation, (c) engaging industry partnership, and (d) incorporating student voice alongside open-access materials. Part III presents the core exercise categories that constitute the redesigned course, illustrating how each assignment requires students to engage with AI while maintaining human judgment as the ultimate measure of competence. Part IV responds to anticipated objections—that this approach won’t prepare students for the bar exam; that students will not learn foundational research, writing, and analytical skills; that students will use AI to cheat; and that AI cannot teach critical thinking—demonstrating how thoughtful design addresses each concern. Part V concludes with reflections on the evolving role of the professor and the institutional imperative to embrace mission-driven AI adoption.
I. THE “WHY”: THE PEDAGOGICAL RATIONALE FOR EMBEDDING AI IN THE LAW SCHOOL CLASSROOM
A. INTRODUCTION
The decision to go “all in” emerged from a convergence of professional experience, personal strengths, and institutional mission. But conviction and mission alignment, while necessary, aren’t enough. The transformation of a foundational and required law school class must rest on sound pedagogical rationale and be supported by evidence that the approach enables students to better fulfill the course learning outcomes.
This Article makes a distinctive contribution to the conversation. Previous legal education scholarship has largely addressed AI in three contexts: as an add-on to existing required courses (supplemental exercises without redesigning core pedagogy), as a separate elective or as mandatory training (standalone course or training about AI and law or about AI and legal practice), or as a compliance matter (policies focused on detecting and preventing unauthorized AI use). This Article describes something fundamentally different: the comprehensive transformation of a required, bar-tested, doctrinal course—first-year Contracts—with AI fully embedded in its design from the ground up, rather than retrofitted as a supplement. It is not theory. It is practice.
The redesigned course rests on a fundamental premise: AI functions as an enhancer of human learning and legal practice, not as a replacement for human judgment or effort. AI amplifies what students can achieve—expanding practice opportunities, accelerating feedback cycles, personalizing learning pathways, and building metacognitive skills—while faculty guidance ensures students develop the critical thinking and professional judgment that remain distinctly human responsibilities. Three interconnected justifications support this enhancement model.
B. REASON I: CREATING ACCESS AND EQUITY THROUGH AI-ENHANCED PEDAGOGY
The most compelling justification for redesigning Contracts rests on creating a more equitable and accessible learning environment that aligns with MHSL’s mission.
1. Neurodiversity, Disabilities, and the Critique of Traditional Pedagogy
Prompted by a colleague’s casual observation that AI could improve the learning of neurodiverse students, I dove into a literature search that validated not just my personal experience as a neurodivergent professor but a systemic problem in legal education. A 2023 Bloomberg Law survey found that 24.7 percent of law students self-identify as neurodivergent—significantly higher than the 6.6 percent of practicing lawyers who report the same.25 This student figure likely understates the actual population, as many students do not disclose their disability status to their law schools.26 The law school population also includes a significant number of students with non-visible disabilities, such as traumatic brain injury, chronic health conditions, and learning disabilities that may or may not require formal accommodations.27 In law school, the disabled and neurodivergent law student population is deeply committed and highly engaged, often spending significantly more time preparing for class than their peers, yet they report substantially lower levels of institutional support and sense of belonging because law school environments are often not designed with them in mind.18 Recent national survey data confirms this pattern: disabled law students are “among the most engaged participants in legal education,” yet they consistently report lower satisfaction with their law school experience (75 percent vs. 84 percent for non-disabled students), less comfort and sense of belonging, and significant gaps in academic and career
25 See BLOOMBERG LAW: LAW SCHOOL PREPAREDNESS 1, 10 (Fall 2023) (Bloomberg Law surveyed more than 2,700 individuals, of which 31 percent were law students and 57 percent were practicing lawyers.).
26 See Jingwei Zhang, One Site for All: Using Universal Design Principles to Create an Inclusive Law Library Website for Neurodivergent Students, 44 LEGAL REFERENCE SERVS. Q. 177, 179 (2025). The actual number of neurodivergent law students is difficult to trace because many students choose not to disclose their disability status to their law schools. Non-disclosure stems from multiple factors: some neurodivergent students may not be aware of the concept of “neurodiversity” itself, while others fear the stigma associated with self-identification as neurodivergent. See HALEY MOSS, GREAT MINDS THINK DIFFERENTLY: NEURODIVERSITY FOR LAWYERS AND OTHER PROFESSIONALS 9 (2021) (discussing barriers to disclosure in legal education and practice).
27 JACQUELYN PETZOLD ET AL., LSSSE 2025 ANNUAL REPORT: DISABILITY IN LAW SCHOOL 5, 8– 9 (2025) (reporting that approximately 20 percent of all law students have a disability or condition that impacts their major life activities, with 83 percent of that population having mental health or developmental disabilities. Specific to non-visible disabilities, 26 percent reported having a chronic medical condition, 9.4 percent had learning disabilities, and 2.7 percent had traumatic or acquired brain injury).
28 See id. at 5 (finding disabled law students “engage at equal or greater levels” than their non- disabled peers who spend more time preparing for class and participate at highest levels, yet report substantially lower levels of institutional support and belonging); Kelly, Be Curious, supra note 19, at 255–56 (describing how neurodiverse law students spend more time preparing for class and developing and deploying strategies that enable them to be successful in the law school classroom, often “over-producing” compared to their neurotypical peers); LSSSE Disabled Law Students See Wide Support Gaps Across Legal Education, NAT’L JURIST (Nov. 14, 2025), https://nationaljurist.com/lssse-disabled-law-students-see-wide-support-gaps-across-legal-education/ (reporting that disabled students “‘do so much on their own—preparing for class, contributing to discussions, joining and leading student organizations and other activities at the highest levels,’ but are ‘missing necessary institutional support’”).
and support despite their high levels of participation and preparation.29 Disabled and neurodivergent students face multiple barriers, including navigating complex accommodation processes, combating stigma when advocating for accessible education, and confronting pan-disablist approaches that treat all disabled and neurodivergent students identically despite varying diagnoses and needs.30
The challenge extends beyond those with formal accommodations. Many students remain undiagnosed, refuse to register for accommodations because of persistent stigma, or experience temporary executive function difficulties from stress, burnout, pregnancy, or other life circumstances.31 Executive function challenges—including difficulties with planning and organization, working memory, task initiation and completion, emotional regulation, impulse control, and flexible thinking—affect learning outcomes across the student body, not just those with documented disabilities.32
The foundational methodology of law school, the Socratic Method, often exacerbates these challenges.33 The public, on-demand nature of cold-calling creates fear and anxiety, forcing students to expend immense cognitive and emotional energy on masking their differences.20 This
29PETZOLD ET AL., supra note 27, at 3, 16 (reporting 75 percent of disabled students vs. 84 percent of non-disabled students rate their experience as good or excellent; disabled students are less likely to feel comfortable, valued, or part of law school community, with significant gaps in satisfaction with personal counseling, career advising, and job search support).
30See Katherine Silver Kelly, Building Scaffolding Instead of Barriers: Accessibility for Neurodivergent Law Students, 49 S. ILL. U. L.J. 149, 149 (2024) (describing how law schools unintentionally create institutional barriers and approach accommodations with “only tool is a hammer” mentality) [hereinafter Building Scaffolding]; Rachel Lewis & Rebekah Smith, A Lot Left to Learn, NAT’L MAG. (Oct. 11, 2024), https://nationalmagazine.ca/en-ca/articles/thepractice/young-lawyers/2024/a-lot-left-to-learn (describing pan-disablism in legal education where schools treat all neurodivergent students identically despite varying needs, and noting students are often “made to feel like they’re trying to take advantage of the system” when advocating for accommodations).
31 See ASSOCIATION AM. L. SCHS., AI and Neurodiverse Students (YouTube, July 18, 2024), https://www.youtube.com/watch?v=FWc8-Xf3RJ4 [hereinafter AI and Neurodiverse Students] (noting that persistent stigma associated with neurodiversity causes many students to refuse accommodations, or to even notice accommodations, and that students without accommodations may still struggle with learning); see also PETZOLD ET AL., supra note 27, at 8 (finding 20 percent of law students report having a disability in 2026, yet many do not seek formal accommodations).
32 See AI and Neurodiverse Students, supra note 31 (identifying executive function challenges including planning and organization, working memory, task initiation and completion, emotional regulation, impulse control, and flexible thinking as affecting neurodivergent students and others experiencing temporary cognitive difficulties from stress, burnout, or life circumstances).
33 See Duhl, Over the Borderline, supra note 16, at 779–80; see also Elizabeth G. Porter, The Socratic Method, in BUILDING ON BEST PRACTICES: TRANSFORMING LEGAL EDUCATION IN A CHANGING WORLD 101 (Deborah Maranville et al. eds., 2015) (explaining that “Socratic learning requires students to think on the spot, answer precisely, and take intellectual risks”).
34 Be Curious, supra note 19, at 248.
turns the classroom into a “place of judgment, not curiosity,” especially for neurodivergent learners.35 Even faculty who use a soft Socratic Method (letting students pass if they are unprepared, giving students advance notice when they will be on call, or calling only on students who volunteer) can create anxiety or discomfort for students who choose to participate, and feelings of isolation and incompetence for students who do not. My new course design directly counters this in three ways.
First, by making learning individualized, such as by moving the Socratic dialogue into a private, one-on-one exchange between a student and a Socratic Bot programmed to inquire into the assigned materials, students practice articulating legal reasoning without the pressure of performing in front of their peers, which eliminates the need for masking and provides the psychological safety required for deeper processing.21 These individualized dialogues can then be reviewed by the professor, who can make targeted interventions of students who are struggling with comprehension or analysis.
Second, for challenges related to executive function, AI provides critical cognitive scaffolding by assisting with structuring ideas, distilling complex texts to relieve cognitive overload, and improving organization. Empirical research on AI’s cognitive effects among legal professionals confirms these benefits: AI reduces emotional strain by 16 percent in client-intake tasks, reduces active mental focus by 72 percent in calculating billable hours, and reduces memory demand by 11 percent in document review like reviewing and summarizing a will, with an overall cognitive load reduction of up to 25 percent. Significantly, students using AI were 129 percent more likely to answer comprehension questions correctly and 40 percent more likely to complete tasks at all.37 For example, AI can help students break large writing assignments into manageable tasks with interim deadlines, transform dense judicial opinions into organized outlines highlighting key holdings and reasoning, generate customized study aids that present information through visual charts, timelines, or concept maps tailored to individual learning preferences, or generate hypotheticals to which students can provide answers and get feedback. This scaffolding creates a judgment-free space to learn and practice, particularly when a student might be uncomfortable seeking the assistance of a professor.38
35 Id. at 245.
36 See generally Harrington, supra note 2; see also Be Curious, supra note 19, at 259–63 (discussing masking and how participation in the law school setting often leads to masking), 295–96 (discussing how the traditional Socratic method requires a lot of extra energy for neurodivergent students and providing recommendations for implementing this method in ways that increase accessibility).
37 THEMIS SOLUTIONS INC., AI-DRIVEN GROWTH: 2025 LEGAL TRENDS REPORT 44, 46, 48, 52 (2025), https://www.clio.com/resources/legal-trends/ (last accessed Dec. 13, 2025) (reporting results from Neuro-Insight SST study measuring cognitive load reduction across multiple legal tasks); On scaffolding, See AI and Neurodiverse Students, supra note 31 (providing examples of AI cognitive scaffolding including breaking large assignments into manageable tasks with interim deadlines using tools like Goblin Tools’ “Magic ToDo” tool, organizing complex information into visual formats, and providing structure for students experiencing executive function challenges); Kelly, Building Scaffolding, supra note 30, at 294–95 (discussing scaffolding for neurodivergent students). Goblin Tools’ “Magic ToDo” is available in the Apple app store at https://apps.apple.com/us/app/goblin-tools/id6449003064.
Third, implementing AI for accessibility as embedded design instead of mere accommodation creates benefits for all students—a phenomenon known as the “curb-cut effect,” which aligns with the Universal Design for Learning (UDL) framework, which operates from the understanding that when we create an accessible course, it benefits everyone.39 This term originates from the physical curb cuts in sidewalks: originally designed to provide wheelchair users access to sidewalks, these cuts turned out to benefit everyone: parents with strollers, travelers with rolling luggage, delivery workers with carts, and cyclists all gained easier passage. Similarly, when we design AI integration with neurodivergent and other disabled students’ needs in mind from the outset rather than adding accommodations as an afterthought, the resulting flexibility benefits the entire class of students.40 When the classroom environment allows multiple modes of practice—with some students using AI to convert cases to audio while others are using it to generate visual concept maps, some receiving grammar feedback while others are getting help organizing research notes—all students flourish.41 Educational research demonstrates that AI can level the playing field in powerful ways: AI tools provide students with additional academic support previously reserved for those with resources to hire private tutors, including text-to-speech and speech-to-text capabilities for students who process information better aurally than visually, instant feedback on practice problems allowing for self-paced mastery, and grammar and spelling assistance that reduces barriers for students with dyslexia or for whom English is not their first language.42
38 See AI and Neurodiverse Students, supra note 31 (demonstrating practical AI accessibility implementations including AI-generated practice hypotheticals where students answer questions and receive feedback on their reasoning; text-to-speech readers that convert case PDFs to audio for students who process information better aurally; AI-powered outline generators that distill complex readings into organized summaries; and grammar/spelling assistance that provides patient, iterative feedback on writing mechanics without judgment).
39Kelly, Be Curious, supra note 19, at 275 (discussing the “curb-cut effect” where accommodations benefit all); see also generally DAVID H. ROSE & ANNE MEYER, TEACHING EVERY STUDENT IN THE DIGITAL AGE: UNIVERSAL DESIGN FOR LEARNING (2002) (foundational text on UDL principles).
40 Kelly, Be Curious, supra note 19, at 275 (discussing the “curb-cut effect” where accommodations benefit all).
41 See AI and Neurodiverse Students, supra note 31 (explaining Universal Design for Learning principles: multiple means of presentation allow presenting content through text, audio, visual charts, timelines, and concept maps; multiple means of engagement include varied methods for student motivation and participation; multiple means of demonstrating learning enable students to show mastery through different formats); see also generally Rose & Meyer, supra note 39 (foundational text on UDL principles).
42 See AI and Neurodiverse Students, supra note 31 (documenting how AI provides previously inaccessible academic support including text-to-speech and speech-to-text for multimodal learning, instant feedback for self-paced practice, organizational tools for time management and task breakdown, and writing assistance for grammar and spelling); Hoang Pham et al., How Will AI Impact Racial Disparities in Education?, STAN. L. SCH.: BLOG (June 29, 2024), https://law.stanford.edu/2024/06/29/how-will-ai-impact-racial-disparities-in-education/ (finding AI-enabled personalized learning systems helped underserved students achieve “nearly double the gains” on standardized assessments); Lorna Gonzalez et al., Leveraging Generative AI for Inclusive Excellence in Higher Education, EDUCAUSE REV. (Aug. 15, 2024), https://er.educause.edu/articles/2024/8/leveraging-generative-ai-for-inclusive-excellence-in-higher-education (AI can be leveraged for inclusive excellence in higher education by supporting mission-focused practices in three key areas: accessibility, identity, and epistemology); Heather Nester, How First-Generation College Students Can Use AI to Level the Job-Search Playing Field, NACE (Aug. 22, 2024), https://www.naceweb.org/diversity-equity-and-inclusion/best-practices/how-first-generation-college-students-can-use-ai-to-level-the-job-search-playing-field (first-generation college students use of AI to research careers and gain job-search support when faced with the disadvantage of a smaller professional network compared to their peers).
One of my goals for the course is to teach students how to build and customize these academic support tools themselves—using AI to generate practice multiple-choice questions or essay hypotheticals on the material, answer those questions or hypotheticals, and then receive feedback on their answers—thereby helping bridge the digital divide between students with different levels of technological literacy and access.43
2. Closing Resource Gaps for Underrepresented Students
The course’s mission is fundamentally tied to access and equity. AI acts as an equalizer that can level the playing field for first-generation, low-income, and underrepresented students by democratizing resources previously reserved for the privileged.44 AI tools serve as an always available, personalized tutor and study aid, offering 24/7 support that affluent students might purchase privately. These tools support study habits through self-quizzing and explaining complex concepts in a judgment-free space. Empirical studies confirm that personalized learning systems driven by AI have helped under served students achieve “nearly double the gains” on standardized assessments, validating AI’s potential to close resource gaps and strengthen learning outcomes.45
AI-enabled personalized learning systems helped under served students achieve “nearly double the gains” on standardized assessments); Lorna Gonzalez et al., Leveraging Generative AI for Inclusive Excellence in Higher Education, EDUCAUSE REV. (Aug. 15, 2024), https://er.educause.edu/articles/2024/8/leveraging-generative-ai-for-inclusive-excellence-inhigher-education (AI can be leveraged for inclusive excellence in higher education by supporting mission-focused practices in three key areas: accessibility, identity, and epistemology); Heather Nester, How First-Generation College Students Can Use AI to Level the Job-Search Playing Field, NACE (Aug. 22, 2024), https://www.naceweb.org/diversity-equity-and-inclusion/bestpractices/how-first-generation-college-students-can-use-ai-to-level-the-job-search-playing-field (first-generation college students use of AI to research careers and gain job-search support when faced with the disadvantage of a smaller professional network compared to their peers). See AI and Neurodiverse Students, supra note 31 (providing example of teaching students to use AI to generate hypothetical problems on course material, formulate answers, and receive constructive feedback—a process that develops both substantive knowledge and technological competence); see also Gergen, supra note 12 (discussing one law student’s experience using AI to study and prepare for the bar exam).
43 See AI and Neurodiverse Students, supra note 31 (providing example of teaching students to use AI to generate hypothetical problems on course material, formulate answers, and receive constructive feedback—a process that develops both substantive knowledge and technological competence); see also Gergen, supra note 12 (discussing one law student’s experience using AI to study and prepare for the bar exam).
44 See AI and Neurodiverse Students, supra note 31 (documenting how AI provides previously inaccessible academic support including text-to-speech and speech-to-text for multimodal learning, instant feedback for self-paced practice, organizational tools for time management and task breakdown, and writing assistance for grammar and spelling).
45 See Pham et al., supra note 41 (finding AI-enabled personalized learning systems helped under served students achieve “nearly double the gains” on standardized assessments). AI democratizes resources previously reserved for the privileged by providing 24/7 personalized tutoring, self-quizzing tools, and judgment-free explanations of complex concepts—support that affluent students might purchase privately but that AI makes accessible to first-generation and underrepresented students. See also Gergen, supra note 12 (on one law student’s experience using AI to study and prepare for bar exam).
Other law schools are similarly recognizing AI’s potential to address access to justice while strengthening legal pedagogy. For example, Vanderbilt Law School’s AI Law Lab works to extend legal services and bridge the justice gap through AI-integrated curriculum, demonstrating that institutions across the country view AI as an opportunity to advance both educational excellence and equitable access to legal services.46
3. Enhancing Learning Outcomes: Repetition and Feedback
The redesigned Contracts course leverages two powerful, empirically validated principles of learning science that traditional law school classes often neglect: frequent practice and immediate feedback. First, traditional legal pedagogy limits substantive student practice, whereas by integrating AI tools, every student receives greatly increased individualized learning opportunities—for example, using AI to generate hypotheticals on the material, formulating answers to those problems, and receiving immediate feedback on their reasoning.47 This cycle of generation, application, and feedback can be repeated until mastery is achieved, leading to better long-term retention of doctrinal knowledge.48
C. REASON 2: PREPARING PRACTICE-READY LAWYERS
The second justification is professional: law schools have a duty to prepare graduates for the legal profession as it is undergoing transformation. We must ask, “Are we preparing students for a practice that is rapidly evolving with AI, or for a practice model that existed months or years ago?”
1. The Reality of Legal Practice and the Competence Gap
46 Vanderbilt AI Law Lab, VAND. L. SCH., https://law.vanderbilt.edu/vanderbilt-ai-law-lab/ (last visited Dec. 3, 2025).
47 See Johana Fleckenstein et al., Automated Feedback and Writing: A Multi-Level Meta-Analysis of Effects on Students’ Performance, in FRONTIERS: FRONTIERS IN ARTIFICIAL INTELLIGENCE (July 3, 2023), https://www.frontiersin.org/journals/artificialintelligence/articles/10.3389/frai.2023.1162454/full (finding AI is significant for learning transfer tasks); Zhang et al., supra note 2, at 1 (finding hybrid feedback [both AI and human] superior for higher-order skills).
48 See AI and Neurodiverse Students, supra note 31 (demonstrating how students use AI to generate practice hypotheticals, answer them, and receive constructive feedback); Fleckenstein et al., supra note 46 (finding significant positive effects of automated feedback on student learning); Aderonke Adegbite & Salami Suleiman, AI-Powered Personalized Learning in Legal Education: A Tool for Developing Future Ready Lawyers, 5 INT’L J. LAW: JUST. & JURIS. 323, 323 (2025) (AI in legal education simulates complex, real-world legal scenarios and provides real-time feedback on analytical processes and simulates complex, real-world legal scenarios); Shai Farber, Harmonizing AI and Human Instruction in Legal Education: A Case Study from Israel on Training Future Legal Professionals, 31 INT’L J. LEGAL PRO. 349, 352–353 (2024) (AI personalized learning systems, such as adaptive learning and intelligent tutoring, dynamically adjusts content based on a student’s success and provides immediate feedback on the student’s analytical processes.).
The legal profession is being fundamentally transformed by AI. Law firms, courts, and corporate legal departments are integrating AI tools for tasks such as practice management, legal research, cite-checking, e-discovery, and contract review, meaning legal practice now requires new “metaskills,” particularly AI literacy.49 For lawyers, AI literacy is defined as the ability to prompt effectively and iterate on nuanced inputs; verify the accuracy of AI outputs (checking for hallucinations); identify bias and ethical issues in AI-generated content; and supervise the AI output, treating it as a specialized co-collaborator.50 My redesigned Contracts course ensures students begin to acquire fluency and comfort in AI literacy and provides them the opportunity to use professional tools such as Spellbook (for contract drafting, revision, and negotiation) and Lexis+ AI (for legal research and writing). We want students to learn how to use AI while in the safe environment of law school rather than when the stakes are higher once they graduate and join a law firm or take on another job as a lawyer.
2. The Essential Role of Human Judgment
While AI excels at automating routine, process-driven work (the lower levels of Bloom’s taxonomy), it cannot (yet) perform the functions that define a lawyer’s highest value. AI does not replicate empathy in client relationships; engage in complex problem-solving, dispute resolution, or development of novel legal arguments; or exercise ethical judgment and strategy (e.g., negotiation strategy, strategic litigation decisions).51 The goal of this course is for students to appreciate what tasks can be transferred to the machine so that the student—the human overseeing AI—can dedicate their focus and energy to higher-order skills like evaluation, analysis, and judgment.
49 See Jen Leonard & Bridget McCormack, How AI Is Changing Legal Education with Dyane O’Leary and Jonah Perlin, AM. ARB. ASSOCIATION (Oct. 14, 2025), https://www.adr.org/podcasts/ai-and-the-future-of-law/how-ai-is-changing-legal-education-withdyane-o-leary-and-jonah-perlin; John Lande, Solving Professors’ Dilemmas About Prohibiting or Promoting Student AI Use. Rather Than Avoiding AI 6 (U. Mo. Sch. Legal Studies, Working Paper No. 2025-53, Dec. 1, 2025), https://ssrn.com/abstract=5841522 (law school faculty have a critical role helping “students build the judgment, creativity, and ethical awareness they will need to use AI tools in practice. Law school is a particularly important place to teach responsible AI use, partly because faculty can supervise students in ways that employers may not advocates for.”).
50 See David S. Kemp, Artificial Intelligence for Lawyers and Law Students: Crutch, Craft, or Catalyst?, 49 SETON HALL J. LEGIS. PUB. POL’Y 633, 638 (2025); see also Samantha A. Moppett, Preparing Students for the Artificial Intelligence Era: The Crucial Role of Critical Thinking Skills, 52 MITCHELL HAMLINE L. REV. 240, 257 (2025); See also Ryan Harroff & Jon Campisi, AI Boom Forces Law Firm Tech Leaders to Rethink Training Practices, AM LAW (Dec. 4, 2025) (reporting that law firms find AI training more complex than previous technologies due to constant updates, requiring continuous rather than one-time training sessions, with some firms dedicating up to 20 percent of junior attorney billable time to AI skill development).
51 Kemp, supra note 49, at 642 (listing skills AI cannot perform, such as negotiation and client consultation); Moppett, supra note 49, at 267 (listing empathy, judgment, and strategic advice).
This approach addresses a critical concern about AI’s impact on the legal profession: the risk of eliminating entry-level jobs.52 AI tools like Spellbook can draft contracts more efficiently and effectively than most (if not all) young associates, but those contracts still require review by senior lawyers. If firms replace junior positions with AI, where will the next generation of senior lawyers come from? The succession gap threatens the profession’s future. As legal scholar Dana Remus observed over a decade ago when early automation technologies began transforming document review: “Recent law school graduates may lament the impact on entry-level law hiring.”53 This concern has only intensified with generative AI, which automates not just document review but legal research, contract drafting, and other foundational tasks that traditionally provided junior lawyers with billable work and critical skill development.54 Recent empirical research confirms this trajectory: organizations that adopt generative AI reduce junior hiring by 7 to 12 percent (a percentage that is likely to grow as AI becomes more sophisticated), while continuing to expand senior roles, with the adjustment coming primarily from hiring fewer new juniors rather than from layoffs.55
Legal education must therefore ensure that students develop fundamental skills through AI use rather than having AI substitute for that learning. When students use Spellbook in this course, they are not outsourcing contract drafting; rather, they are practicing the iterative process of drafting, reviewing, and revising with AI as a collaborator. They learn to spot issues in AI generated language, understand strategic choices in contract terms, and exercise professional judgment over the final product. This prepares them to step into roles where they can competently supervise AI tools from day one, rather than being rendered redundant by them.
52 See Claire Beresford, The Impact of AI and Legal Tech on Junior Lawyers: Risks, Opportunities, and Career Strategies, LAURENCE SIMONS (2024), https://laurencesimons.com/news-insights/the-impact-of-ai-and-legal-tech-on-junior-lawyersrisks-opportunities-and-career-strategies/ (last visited Dec. 2, 2025) (documenting how AI reduces junior lawyers’ hands-on experience with legal research, document review, and drafting that historically built foundational skills); Ken Crutchfield & Jennifer McIver, Straight Talk: How Will AI Impact the Next Generation of Lawyers?, WOLTERS KLUWER (July 9, 2025), https://www.wolterskluwer.com/en/expert-insights/straight-talk-how-will-ai-impact-the-nextgeneration-of-lawyers (noting “young associates are not going to be able to rely on the time intensive, mundane tasks to achieve higher billable hours; rather, they’re going to have to continually fill their billables with more mentally challenging and strategic work”).
53 See Dana Remus, The Uncertain Promise of Predictive Coding, 99 IOWA L. REV. 1691, 1691 (2014) (noting that machine-learning technologies automating document review prompted concerns about “the impact on entry-level law hiring”); see also Dana Remus & Frank S. Levy, Can Robots Be Lawyers? Computers, Lawyers, and the Practice of Law (Nov. 30, 2016) (unpublished manuscript), https://ssrn.com/abstract=2701092 (analyzing automation’s impact on demand for lawyers’ time and discussing “the ways in which computers are changing—not simply replacing—the work of lawyers”).
54 See Beresford, supra note 51; Crutchfield & McIver, supra note 51.
55 See Guy Lichtinger & Seyed Mahdi Hosseini Maasoum, Generative AI as Seniority-Biased Technological Change: Evidence from U.S. Résumé and Job Posting Data 4 (Nov. 11, 2025) (unpublished manuscript), https://ssrn.com/abstract=5425555 (finding that six quarters after AI adoption, junior headcount is 9 percent lower than comparable non-adopting firms, while senior headcount continues rising).
The solution is not to avoid AI adoption, but to manage the transition thoughtfully. Firms must hire young professionals who are well-versed in AI, capable of using it to enhance their work while understanding its limitations and receiving mentorship from senior lawyers. With proper training, AI-equipped junior lawyers will be more capable than their predecessors: able to handle greater complexity earlier in their careers because routine tasks are automated, freeing them to focus on higher-order skills from the outset.
For example, AI can make legal research more efficient. But lawyers must learn the importance of verifying AI output, as courts have already sanctioned lawyers who failed to check hallucinated cases.56 Judge Robert McBurney of the Superior Court of Fulton County, Georgia, a law school classmate of mine, recently commented, AI is a “great launchpad” for research, but “[i]t is not yet ready for prime time when it comes to the finished product.”57 This mandates that students learn how to use AI as a launchpad but also the human judgment and skill, including the verification of sources, to create a final product.
D. REASON 3: SHIFTING THE NARRATIVE FROM POLICING TO PEDAGOGY
The dominant narrative around AI in higher education has been one of fear: How do we prevent students from cheating? How do we detect AI-generated work? How do we preserve academic integrity? This narrative is both unproductive and actively harmful.
1. The Harms of Surveillance and Distrust
The prevailing narrative focuses on deploying remote proctoring software and AI detection tools to police students. This approach is fundamentally flawed and destructive to learning because of three core harms. First, the detection tools are unreliable: they have been empirically proven to be highly inconsistent and prone to false positives (incorrectly flagging human-written work as AI-generated), rendering them unreliable as the sole determinant in academic integrity cases.58
56 See, e.g., Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 448 (S.D.N.Y. 2023) (sanctioning attorneys $5,000 for submitting brief citing nonexistent cases generated by ChatGPT, holding that while “there is nothing inherently improper about using a reliable artificial intelligence tool for assistance,” attorneys have a “gatekeeping role” to “ensure the accuracy of their filings”); Gauthier v. Goodyear Tire & Rubber Co., No. 1:23-CV-281, 2024 WL 4882651, at *3 (E.D. Tex. Nov. 19, 2024) (sanctioning attorney $2,000 and requiring CLE course for submitting brief with AI-generated hallucinated cases, emphasizing failure to verify AI output breached ethical obligations under Rule 11); see also Christopher F. Lyon, Fake Cases, Real Consequences: Misuse of ChatGPT, 28 NYSBA N.Y. LITIGATOR, No. 2, 2023, at 12 (analyzing Mata sanctions and noting that while judge did not fault attorneys for using ChatGPT, “critical points that resulted in sanctions” were attorneys’ failure to verify accuracy and continued reliance on fake cases despite warnings).
57 See E-mail from Robert McBurney, Judge, Superior Ct. of Fulton Cnty., Ga., to Gregory Duhl, Professor of Law, Mitchell Hamline Sch. of L. (Nov. 9, 2025) (on file with author).
58 See Ahmed M. Elkhatat et al., Evaluating the Efficacy of AI Content Detection Tools in Differentiating Between Human and AI-Generated Text, 19 INT’L J. EDUC. INTEGRITY 1, 8 (2023) (reporting inconsistency and high false positive rates); Doraid Dalalah & Osama M.A. Dalalah, The False Positives and False Negatives of Generative AI Detection Tools in Education and Academic Research, 21 INT’L J. MGMT. EDUC. 1,12 (2023).
These false positives create a climate of suspicion where innocent students believe that they could be unfairly accused of cheating. Tools such as remote proctoring software are also unreliable in that the technology has failed, giving rise to anticipatory anxiety among students who are already under the intense stress of taking final exams.59
Second, and most fundamental, AI detection tools erode the faculty–student trust that is the cornerstone of effective learning.60 When students trust their professors, they become more vulnerable, more receptive to feedback, and more willing to take intellectual risks—all essential conditions for deep learning.61 Surveillance-based approaches establish what scholars describe as a “low-trust environment” and a “profoundly unsettling culture of distrust” between faculty and students, leading students to feel unsafe asking questions or taking risks, exploring AI use, or disclosing when they have used it.34 This erosion of trust undermines the pedagogical relationship that makes learning possible.
Finally, AI detection tools carry equity harms by being invasive of student privacy and exhibiting algorithmic bias that could unfairly flag students with disabilities, students of color, or non-native speakers, penalizing the students whom AI aims to support.63 Even when students are The False Positives and False Negatives of Generative AI Detection Tools in Education and Academic
59 See Meazure Learning Class Action, CLASSACTION 101 (Oct. 24, 2025), https://classaction101.com/lawsuits/meazure-learning-class-action/ (class action lawsuit filed by California bar exam test-takers affected by platform failures of remote proctor vendor Meazure Learning during February 2025 exam).
60 See Jiahui Luo, How Does GenAI Affect Trust in Teacher-Student Relationships?, 30 TEACHING IN HIGHER EDUC. 991, 991–92 (2025).
61 See Luo, supra note 59, at 992 (documenting how trust enables students’ willingness to be vulnerable, take risks, and engage authentically with learning); Patricia Williams, Opinion Forum: How AI Is Changing Higher Education, CHRON. HIGHER EDUC., https://www.chronicle.com/article/how-ai-is-changing-higher-education (Nov. 5, 2025) (noting that trust-based relationships allow students to “be vulnerable in asking questions” and “receptive to constructive feedback”).
62 See Williams, supra note 60 (describing “profoundly unsettling culture of distrust”); see also
Clay Shirky, The Post-Plagiarism University, CHRON. HIGHER EDUC. (Nov. 3, 2025), https://www.chronicle.com/article/the-post-plagiarism-university (arguing surveillance approach creates adversarial rather than collaborative learning environment).
63 See Symposium, Examining the Examiners: Students’ Privacy and Security Perceptions of Online Proctoring Services, 17 USENIX 633, 634 (2021) (finding 61 percent could still cheat but were worried about surveillance); Samantha Mita, AI Proctoring: Academic Integrity vs. Student Rights, 74 HASTINGS L.J. 1513, 1553 (2022) (noting AI proctoring risks overlooking legal and ethical consequences); Monika Blue Kwapisz et al., Surveillance and Disability in Online Proctored Exams: Student Perspectives and Design Implications 1 (Nov. 13, 2025) (unpublished manuscript), https://arxiv.org/pdf/2511.10826v1 (AI proctoring violates privacy by being overly intrusive of student’s personal space, by depriving student’s ability to meaningfully consent to data being shared, by concerns of where the data is shared and how long it is stored, and by risk of hacks and leaks of personal data); Addie Griffey, Legal Implications of Remote Exam Proctoring: Extending Federal Law to Protect Student Privacy, 74 CASE WESTERN L. REV. 791. 825 (2024) (mandatory room scans for remote proctoring may violate the Fourth Amendment protection against unreasonable search by exposing a student’s private bedroom or other space and biometric data vulnerable to security breaches); see also Bonnie Stewart, Online Exam Monitoring Can Invade Privacy and Erode Trust at Universities, THE CONVERSATION (Dec. 3, 2020), https://theconversation.com/online-exam-monitoring-can-invade-privacy-and-erode-trust-at-universities-149335 (“Testing and proctoring methods that invade privacy and erode trust end up undermining the very integrity that institutions demand students uphold.”).
not unfairly flagged, remote proctoring, for example, intrudes into the student’s private space and could cause discomfort to a student with tics or who stims, or who lives in a crowded apartment with lots of noise in the unit, or who has difficulty performing or experiences paranoia when being watched. A pedagogy built on surveillance and distrust interferes with our role as facilitators of the learning process.
2. The Alternative: Design, Don’t Police
This pattern of technological resistance is not new to education. Calculators were once banned from mathematics classrooms as cheating tools; now they are standard equipment. Graphing calculators, word processors, the internet, online learning—each innovation faced similar resistance before becoming integrated into effective pedagogy. Over time, the initial resistance and skepticism usually give way to acceptance and eventually active engagement with the new technology.
Today, there is a better approach than policing: design assignments where AI enhances learning rather than undermines it. Professor Clay Shirky argues that trying to fit AI use into traditional definitions of plagiarism fails because “first, generative AI is not a human author, and second, the students know it”; asserting that many students thus see unauthorized AI use as a “victimless crime.”64
We must change the mindset from “How do I catch cheaters?” to “How do I design learning activities and assignments where AI enhances learning?” The Contracts course is built on the pedagogical imperative of trust. By making the permitted use of AI transparent, explicit, and essential to student success, we transform AI from a means of cheating into a necessary tool of the trade. Students are told: AI use is not only allowed, it is required, but you are responsible for the outcome. As Adam Borrego, an academic advisor at my school observed, responsible students “see AI more as a way to help find relevant information that they can then analyze and process on their own. They see it as a tool rather than as something used to cheat.”65
Surely, students will use it to cheat. But some students used other methods to cheat before AI. There will always be those students. There will be natural consequences for those folks when it comes time to take the bar exam.
The solution is not policing their behavior but designing learning activities and assessments that compel human judgment. For instance, an assessment that requires students to show their improvements over AI-generated work product using Track Changes, or risk a poor grade if they rely on a hallucinated citation in a brief, memo, or judicial opinion they draft and do not check
AI’s research, or if they fail to supplement AI’s output with more fact-specific, nuanced analysis.
This approach also challenges a troubling hierarchy in legal education. Legal writing and research professors have been at the forefront of integrating AI into their teaching, yet they are sometimes viewed as teaching “skills” rather than “real” doctrine (even though legal research and writing professors are at the forefront of teaching students how to analyze doctrine).66 The implicit assumption is that AI is acceptable for practice or skills courses but not for traditional doctrinal courses. This assumption is wrong on two levels. First, AI can enhance doctrinal learning through individualized practice and repetition. For example, a student can apply the consideration doctrine to bot-generated hypotheticals and receive feedback until they fully understand the doctrine. Second, and more fundamentally, the distinction between “skills” and “doctrine” is artificial.67 The best doctrinal teaching has always taught students how to think like lawyers: how to read cases, construct arguments, and apply rules to facts. These are skills, and AI is a powerful tool for practicing them.
The imperative is clear: if we embrace AI for legal education, it demands that professors change the way they teach. We should no longer rely on lectures that students could access through AI or on class-wide Socratic dialogue. We can no longer give take-home exams, seminar papers, and other assignments that AI could complete while telling students that using AI violates the honor code. We must instead design our courses to foster active learning, critical evaluation, and human judgment—the very skills that AI makes more crucial.68
66 See TAMMY PETTINATO OLTZ, LAWYERING SKILLS IN THE DOCTRINAL CLASSROOM xi (2016) (“law schools continue to struggle with an artificial split between ‘doctrinal’ courses and ‘skills’ courses—a split that ignores best practices and undermines student learning”); Linda H. Edwards, Legal Writing: A Doctrinal Course, 1 SAVANNAH L. REV. 1, 4–6, 12 (2013) (arguing legal writing courses teach doctrine through analysis of legal reasoning, case reading, and argument construction, yet are treated as separate from “real” doctrine courses); Jeremiah A. Ho, Function, Form, and Strawberries: Subverting Langdell, 64 J. LEGAL EDUC. 656, 661 n.4 (2015) (recognizing the constructed hierarchy between skills and doctrinal instruction in legal education); Deborah Jones Merritt, Caste Revisited, L. SCH. CAFE (Aug. 13, 2021), https://www.lawschoolcafe.org/2021/08/13/caste-revisited/ (describing “caste system in legal education: a hierarchy that favors professors who teach torts, contracts, and other legal ‘doctrine’ over those who teach legal writing, clinics, and other legal ‘skills’”); see also Larry Cunningham, Dividing Law School Faculties Into Academic Departments: A Potential Solution to the Gendered Doctrinal/Skills Hierarchy in Legal Education, 67 VILL. L. REV. 679, 682 (2022) (documenting how skills faculty “are relegated to a lesser status” despite teaching doctrine through legal reasoning and analysis).
67 See sources cited as supra note 65.
The pedagogical rationale for embedding AI is compelling. But rationale alone is not enough. The question is how to implement this transformation in practice—how to design a Contracts course that integrates AI while maintaining academic rigor, building student competence, and preparing students for both the bar exam and the practice of law.
II. THE “HOW”: IMPLEMENTATION STRATEGY
A. INTRODUCTION
Even when the rationale is compelling, execution matters for good pedagogy. I followed the outcomes-based course design process—commonly called “backward design”—in which I first identified the desired learning outcomes, then determined how to assess whether students achieved those outcomes, and finally designed learning activities aligned with both the outcomes and assessments.69 This approach, which education scholars Grant Wiggins and Jay McTighe systematically developed in their influential Understanding by Design, reverses the traditional course planning sequence in which instructors typically start with content and activities.70 In backward design, every learning activity and assessment flows directly from clearly articulated learning goals, ensuring coherence and intentionality throughout the course.71 L. Dee Fink’s complementary framework of “integrated course design” emphasizes this same alignment principle, demonstrating how effective course design requires systematic integration of learning goals, teaching and learning activities, and feedback and assessment procedures.72
My implementation strategy was built upon four foundational pillars, developing the course as a complex design project that maximized institutional strengths and leveraged external innovation: (1) partnership with in-house instructional designers, (2) crowdsourcing ideas from other faculty, (3) engaging in industry partnerships, and (4) embracing student voice and open-access materials.
B. RESOURCE 1: INSTRUCTIONAL DESIGN PARTNERSHIP
68 See Malcolm Tight, Challenging Cheating in Higher Education: A Review of Research and Practice, 49 ASSESSMENT & EVALUATION IN HIGHER EDUC. 911, 920 (2024) (noting “surprisingly little evidence on what works”).
69 See GRANT WIGGINS & JAY MCTIGHE, UNDERSTANDING BY DESIGN 13–34 (2d ed. 2005) (describing the three-stage backward design process).
70 Id. at 14–15.
71 Id. at 17–22 (explaining how backward design focuses curriculum and teaching on development and deepening of student understanding rather than merely “covering content”).
72 See L. DEE FINK, CREATING SIGNIFICANT LEARNING EXPERIENCES: AN INTEGRATED APPROACH TO DESIGNING COLLEGE COURSES 67–87 (2013) (presenting the integrated course design model emphasizing alignment among situational factors, learning goals, feedback and assessment, and teaching and learning activities).
A primary reason MHSL was able to launch the first ABA-approved hybrid JD program in 2015 and continue to grow the program was the conscious decision to invest in and keep instructional design expertise in-house.73 This foundational lesson—that major curriculum shifts must be driven by pedagogical experts who understand the institutional mission—is also applicable to the Contracts redesign.
My partnership with Amanda Soderlind, the Director of Instructional Design and Development, is the anchor of this transformation. Her role extends beyond the technical or logistical; it is pedagogical. As Amanda explains: “Instructional designers can serve as the bridge between the technology and the pedagogy . . . help[ing] shift [the] mindset from, ‘How can I use AI in my course?’ to more focused on questions like, ‘What new legal reasoning skills or ethical perspectives can students develop because of these tools?’ That shift requires thinking about AI in terms of educational outcomes and considering how it can be used to support critical thinking and reflection rather than basic convenience or automation.”74
Her commitment is essential to maintaining rigor. We are not pursuing AI for the sake of automation or convenience, although those are welcome side effects. Instead, every learning activity or assignment is designed to promote critical thinking and reflection, ensuring that innovation retains the human values at the core of legal reasoning and, more generally, the practice of law. An instructional designer’s support is crucial for translating the theoretical mandate of creating access and preparing practice-ready professionals into concrete experiences for a classroom of students.
B. RESOURCE 2: LEARNING FROM FACULTY INNOVATION
Because the challenge of integrating AI is shared across legal education, I turned to the work of innovative colleagues and academics who have pioneered or experimented with specific uses of AI. I had no formal training in or experience with AI when I started this redesign project. I am not a “techy.” I adapted four proven pedagogical strategies to the Contracts context.
First, the core of the course’s formative assessment, the Socratic Dialogue Bots, was inspired by pioneering work in AI-assisted legal assessment.46 The concept is simple but profound: leverage
73 See generally Janus, supra note 13 (detailing the strategic decision to invest in an in-house instructional design team during the hybrid JD program development). Mitra Madanchian et al., Integrating AI Tools to Enhance Learning Outcomes in Modern Education Systems, 263 PROCEDIA COMPUT. SCI. 514, 517 (2025) (“Professional development initiatives should include thorough instruction on properly using artificial intelligence technologies, including knowledge of their features and advantages. Teachers also depend on constant support as they need tools and resources to remain current with technology developments and handle any issues. Investing in professional development helps educational institutions guarantee that instructors have the tools and expertise required to maximize AI technologies and smoothly incorporate them into their daily work.”).
74 See E-mail from Amanda Soderlind, Director, Instructional Design & Dev., Mitchell Hamline Sch. of L. to Gregory M. Duhl, Professor of Law, Mitchell Hamline Sch. of L. (Nov. 2025) (on file with author).
75 See generally Harrington, supra note 2.
AI to scale the Socratic Method to the entire class to enable each student to engage in a question-and-answer dialogue about the assigned materials. As legal pedagogy research notes, traditional methods of observation (body language, assuming that the speaking students represent the entire class) are vastly inferior to structured Classroom Assessment Techniques for determining students’ understanding.76 I am deploying the bot system after students prepare the materials for the first class in each unit of the course. These opportunities provide six distinct, low-stakes opportunities for students to engage in dialogue for the first fifteen minutes of the class, receive immediate feedback, and upload a transcript for my review. This shifts my role from that of a performer, “knowledge transmitter,” or “questioner,” to a facilitator who analyzes the transcripts to identify patterns of misunderstanding across the class or weaknesses of individual students.77
This pedagogy also builds on the work of my colleague Professor Anthony Niedwiecki who created AI “study buddies” in his Business Organizations course—custom bots for each week’s material that enabled students to complete and receive feedback on practice problems, plus one for the final exam. When I asked a classroom of students for examples of professors at MHSL who integrated AI into their doctrinal courses, several students mentioned how much they relied on Professor Niedwiecki’s “study buddies” to master each week’s material.
Second, I also adapted the concept of bot-building exercises from Professor Niedwiecki, who uses learning-by-teaching as a pedagogical tool.78 Early in the course, I ask students to build a custom chatbot that can teach a law student the consideration doctrine and its exceptions. Drawing on Professor Niedwiecki’s work requiring students to teach “fiduciary duties” to a bot in his Business Organizations course, this exercise requires students to organize knowledge, recognize gaps in their own understanding, and master the doctrine at a deep, structural level.79 By “teaching” the bot, students must think like both instructional designers and subject-matter experts, which deepens their conceptual grasp of contract law fundamentals.
Third, I incorporate AI research tools from colleagues who teach legal research. For example, my colleague Alisha Hennen, an MHSL Research and Instructional Librarian, teaches an advanced legal research course that includes a unit on legal research using AI tools integrated into Bloomberg Law, Lexis+ AI, and Westlaw Precision.
76 Id. at 1 (noting CATs are “vastly superior to our normal methods” of assessing student learning); see also CORNELL UNIV. COMM. REPORT, GENERATIVE ARTIFICIAL INTELLIGENCE FOR EDUCATION AND PEDAGOGY 34 app. G (July 18, 2023) (recommending avoiding unsupervised take-home exams); cf. Classroom Assessment Techniques, U. TEX. AUSTIN, https://ctl.utexas.edu/cats (last visited Dec. 7, 2025) (“Classroom assessment techniques (CATs) are a collection of activities to gather feedback during instruction. Typically, CATs solicit responses from students to provide instructors with information to help them modify their teaching strategies to help students learn more efficiently and effectively.”).
77 Danault, supra note 1, at 1 2 (stating the role shifts from “transmitters of the knowledge” to “facilitators of the learning process”); see also Ghalayini, supra note 23 (describing faculty as the “architects of a new academic era”).
78 See sources cited at supra note 76.
79 See E-mail from Anthony Niedwiecki, Professor of Law, Mitchell Hamline Sch. of L., to Gregory M. Duhl, Professor of Law, Mitchell Hamline Sch. of L. (Nov. 2025) (on file with author).
She teaches prompt generation and source verification—skills that directly informed the design of my judicial opinion research and writing exercise in week 7 of the course. Her research unit demonstrates how AI tools can enhance rather than replace the way students develop traditional legal research competencies, providing a model for integrating AI into doctrinal legal analysis.80 As a general matter, faculty should partner with their library staff when embedding AI into coursework.
Fourth, I incorporate simulation ideas, such as drafting and negotiation using AI, from colleagues who teach dispute resolution and practical skills. For example, my colleague Professor Sharon Press has students mediate disputes between AI-powered bots that simulate different parties, providing low-stakes practice where students can experiment with strategy and tactics without the pressure of live peer evaluation.81 Innovations such as that deployed by Professor Press inspired me to expand AI beyond written research and analysis into experiential contexts where students practice professional skills they’ll need as lawyers.82
C. RESOURCE 3:INDUSTRY PARTNERSHIP WITH SPELLBOOK
To ensure students train on the actual tools they will encounter in practice, Amanda and I pursued an educational partnership with Spellbook, which advertises itself as “the #1 legal AI for transactional lawyers,” using large language models to “review and suggest language for your contracts, right in Microsoft Word.”83 Spellbook provides educational partnerships specifically designed for law schools, offering free access to their enterprise platform for student learning. This step was crucial for grounding the second half of the Contracts course in real-world practice. Internal collaboration and support are critical to integrating external AI tools. Shannon Thomas, MHSL’s chief information officer, has been instrumental in supporting my pedagogical experimentation, including facilitating access to Spellbook, which has made the course design process more collaborative and efficient.
The value proposition of this partnership with Spellbook is straightforward: instead of spending scarce class time teaching students how to become proficient contract drafters by focusing on rote contract formatting and drafting boilerplate and other foundational clauses, the AI handles the routine drafting and compliance checks.84 This frees up students (and me) to focus on higher-order analysis:
80 See E-mail from Alisha Hennen, Research and Instructional Librarian, Mitchell Hamline Sch. of L., to Gregory M. Duhl, Professor of Law, Mitchell Hamline Sch. of L. (Nov. 2025) (on file with author).
81 See E-mail from Sharon Press, Professor of Law, Mitchell Hamline Sch. of L., to Gregory M. Duhl, Professor of Law, Mitchell Hamline Sch. of L. (Nov. 2025) (on file with author).
82 See Gregory M. Duhl & Jaclyn Millner, Transactions and Settlements: Creating a Balance in Legal Education, 14 TENN. J. BUS. L. 517, 520 (2013) (arguing counseling, negotiation, and drafting must be taught as interwoven skills).
83 Legal AI Contract Review & Drafting, SPELLBOOK, https://www.spellbook.legal (last visited Dec. 2, 2025).
84 Evan Absher, Spellbook Is a Unique Offering in the World of Legal AI, LINKEDIN (Sept. 2025), https://www.linkedin.com/posts/evanabsher_spellbook-academic-partnership-program-activity-7368289530417721347-0zmg/ (noting Spellbook allows him to train students on “more meaningful substantive issues” and “less time on being a ‘document maven’”).
recognizing strategic risks, identifying ambiguous contract language, engaging in client communication, and exercising professional judgment—aligning with the Human-in-the-Loop (HITL) model.85 In this approach to AI integration, humans retain supervisory control over automated systems, providing oversight, making context-sensitive decisions, and exercising the ethical and professional judgment that AI cannot replicate.86 The Spellbook exercises progress from using AI to write a first draft of a contract and then students critiquing it, to using AI to review a poorly drafted contract and students evaluating which of Spellbook’s suggested changes to implement, and finally, to live negotiation and contract drafting using the tool, which ensures that students are prepared for the reality of AI-augmented transactional practice.
D. RESOURCE 4: STUDENT VOICE AND OPEN-ACCESS MATERIALS
Our final resource involved direct engagement with students and prospective students. First, students themselves provided valuable insights. Students are early adopters and “digital natives” who crave AI literacy, often providing valuable, unfiltered insights into the technology’s capabilities and limitations.87 As one of our academic advisors noted, many of these highly engaged students see AI as a valuable learning tool instead of a cheating mechanism.88 This perspective helped reinforce the core design principle of trust over policing.
After my realization in Secured Transactions class that Tuesday evening, I spent the first fifteen minutes of office hours each subsequent week brainstorming how to integrate AI into my teaching.89 Students explained how they build AI study tools for their bar-review courses (e.g., AI spits out Multistate Essay Examination and Multistate Performance Test questions and gives the students feedback on their answers). They talked about how they learned about prompt generation and AI legal research tools in their Advanced Legal Research course. They discussed how AI could build learning experiences where students, based on prior performance, are directed to easier tasks (knowledge checks) or harder tasks (practice exercises). Students
85 Human-in-the-loop (HITL) refers to a system in which humans actively participate in the operation, supervision, or decision-making of an automated system to ensure accuracy, safety, accountability, and ethical decision-making. See Rebecca Crootof et al., Humans in the Loop, 75 VAND. L. REV. 429, 438–40 (2023) (defining and analyzing the concept in legal and regulatory contexts).
86 See Cole Stryker, What Is Human in the Loop (HITL)?, IBM, https://www.ibm.com/think/topics/human-in-the-loop (last visited Dec. 2, 2025) (explaining that HITL approaches “help to mitigate the ‘black box’ effect where the reasoning behind AI outputs is unclear” and noting that “even the most advanced deep learning models can struggle with ambiguity, bias or edge cases that deviate from their training data”).
87 See CORNELL UNIV. COMM. REPORT, supra note 75, at 8 (recommending encouraging AI use to “level the playing field for students with disparate abilities and needs”).
88 See Borrego, supra note 64 (on successful students viewing AI as a supplementary tool).
89 In particular, I appreciate the contributions of MHSL students Sharon Ragoonan and Sean Vinsel, who, in addition to contributing to these discussions, also provided me with extensive materials and insights on AI use in legal education and the legal profession.
discussed the tools they use in their jobs. For example, one student who represents creditors in loan disputes has access to vLex, which the company’s website describes as “comprehensive legal intelligence [AI] solutions that transform how legal professionals research, analyze, and practice law across every aspect of their work.”90
Second, market signals from prospective students reinforced the urgency of incorporating AI. MHSL’s vice president of admissions, Annie Gemmell, reported that prospective students are actively asking about AI integration in information sessions, comparing our approach to those of other schools.91 This marketplace pressure aligns with my mission-driven approach, confirming that innovation in AI is no longer a matter of luxury but is rather a necessity for institutional relevance, particularly at an access school like ours.92
Students and prospective students have also bemoaned the high price of law school textbooks.
Therefore, we committed to using an open-access, free casebook for the course, Open Source Contracts: A Free Casebook.93 This choice provides the essential benefit of flexibility, allowing us to structure discrete weekly assignments, integrate specific cases and problems, and feed targeted content into the Socratic Bots without violating the restrictions of traditional copyrighted publisher materials. This choice reduces the students’ financial burden—ensuring equal resource access—and maximizes pedagogical freedom to experiment.
While students do not need to invest in purchasing an expensive textbook, students are required to maintain a $20 per month subscription to ChatGPT Plus throughout the semester, as the custom GPT functionality and advanced features required for the course exercises are only available in the subscription version.94 I’ll encourage students to get the full benefit of their investment by experimenting with other permitted uses for ChatGPT Plus (e.g., building study tools for other classes). A faculty member wishing to experiment with AI in a class should
90 vLEX, https://vlex.com/ (last visited Nov. 30, 2025); vLex, RUTGERS UNIV. LIBRARIES, https://www.libraries.rutgers.edu/databases/vlex (last visited Nov. 30, 2025) (“vLex is a legal research database that contains over 24 million U.S. legal documents, including state and federal case law, statutes, regulations and legislation, administrative codes, specialist and federal court coverage, secondary resources, legal news, and commentary from top law firms.”). vLex has partnerships with over sixty bar associations in the United States, including the Minnesota State Bar Association (of particular relevance to our student body), which give members (including students) free access to the base vLex Fastcase platform. See Bar Associations, VLEX, https://vlex.com/bar-associations (last visited Dec. 13, 2025).
91 See E-mail from Ann Gemmell, Vice President of Admissions, Mitchell Hamline Sch. of L., to Gregory M. Duhl, Professor of Law, Mitchell Hamline Sch. of L. (Nov. 2025) (on file with author).
92 See Ted Underwood, AI Is the Future. Higher Ed Should Shape It., CHRON. HIGHER EDUC. (Nov. 4, 2025), https://www.chronicle.com/article/ai-is-the-future-higher-ed-should-shape-it (arguing that educators must proactively guide AI integration to maintain institutional relevance).
93 See generally MATTHEW BODIE ET AL., OPEN SOURCE CONTRACTS: A FREE CASEBOOK (2025), https://contractscasebook.org/.
94 The free version of ChatGPT does not fully support the custom GPT bots that are central to several course exercises.
explore the extent to which their institutions might buy students enrolled in the class licenses for an applicable AI tool.
These four resources—institutional expertise, faculty innovation, industry tools, and stakeholder student input—informed design choices in the course, from outcomes to the course infrastructure to weekly assignments and learning activities to the final exam format. Such a combination creates a comprehensive implementation strategy that resulted in the six key exercise categories described in the next part. These exercises constitute a scaffolded progression that continually requires the application of human judgment, ethical verification, and sophisticated prompt engineering, building a continuous feedback loop that places the student at the center of the learning process.
III. THE EXERCISES:SHOW AND TELL
A. INTRODUCTION
The transformation of the Contracts course from a traditional, lecture- and discussion-based model to an AI-embedded design is best illustrated through its assignments.95 These six core sets of exercises represent a continuous, scaffolded progression designed to develop AI literacy, doctrinal mastery, and critical professional judgment in parallel. Each assignment follows the pedagogical rationale set forth earlier by requiring students to engage the machine, critique its output, and ultimately assert human expertise. Importantly, these exercises are accompanied by six low-stakes quizzes at the end of each unit to prepare students for a summative closed-book final exam consisting of multiple-choice, past-MBE questions. While AI tools can help students prepare for the bar exam, they must be able to take the bar exam without AI’s help, which is a critical reality this format reinforces.96 These exercises create access and equity (Reason 1) by providing multiple modalities and safe practice spaces for neurodiverse students; prepare practice-ready lawyers (Reason 2) by training them on industry-standard tools; and shift from policing to pedagogy (Reason 3) by making AI use transparent, required, and integral to learning instead of a means of cheating.
A critical design principle underlies all six exercises: AI is assessed on the human value added. Students are never penalized for using AI; they are penalized for using it poorly. Every assignment either requires them to show their prompts, demonstrate their revisions in Track Changes, or prove their independent competence through closed-book components. This assessment philosophy directly implements the “design, don’t police” principle discussed earlier, transforming AI from a potential academic integrity threat into a pedagogical requirement. Table 1 shows an overview of the AI-integrated assignments incorporated into the redesigned course.
95 Faculty interested in adapting these exercises to their own courses may request from me copies of the assignments and rubrics.
96 The quizzes are intentionally low-stakes to reduce anxiety while building the testing stamina and doctrinal recall necessary for high stakes bar-exam-style assessments. This design recognizes that bar preparation requires both AI-assisted learning during law school and the ability to perform without AI assistance during the examination.
Table 1
Overview of AI-Integrated Contracts Course Assignments
| Exercise | Week | Primary AI Tool | Learning Outcome |
| AI Editing for Tone | 1 | Generative AI
(ChatGPT/Claude) |
Construct effective prompts; establish supervisory role over AI |
| Socratic Dialogue Bots | Ongoing (6x) | Custom-trained chatbots (ChatGPT) | Demonstrate proficient doctrinal comprehension and analysis through iterative dialogue |
| Bot Building | 3 | Bot-building platform (ChatGPT) | Achieve mastery-level
understanding by teaching doctrine |
| AI-Assisted Bar Exam Essay | 5 | Generative AI | Generate prompts, critically revise AI output, produce independent work |
| Judicial Opinion | 7 | Lexis+ AI, Westlaw Precision | Verify and evaluate sources, document research process, exercise ethical and human judgment |
| Spellbook Series | Second
half (Weeks 8– 14) |
Spellbook | Identify contract risks and ambiguities, draft and revise contracts with AI support, critically evaluate AI-generated suggestions, recognize when to accept or reject AI recommendations, negotiate with strategic AI support, focus on higher-order analysis while AI handles routine drafting and compliance checks |
B. THE SCAFFOLDING OF AI LITERACY (WEEK 1: EDITING FOR TONE)
The very first assignment introduces students to the fundamental meta-skill of the AI age: prompt engineering. Before using AI for complex legal analysis, students must learn to control the machine’s output precisely.
This introductory, pass–fail assignment, titled “AI Editing for Tone,” asks students to bring a personal biography to class. The lesson focuses on maintaining Content (facts, dates, names) while intentionally manipulating Tone (formality, diction, emotional register). Students receive training on constructing a complex, multi-part prompt—the Setup Block—to direct a generative AI model to perform specific stylistic edits for a target audience (e.g., rewriting the bio in a “Punchy/Marketing” tone or a “Concise/Plain English” tone or a “LinkedIn tone”).
The core objective is immediate and visible: students realize that sophisticated output requires iterative and refined prompts rather than simple commands. This demonstration of control immediately positions the student as the supervisor of the AI, easing their anxiety and establishing a collaborative precedent for the entire semester. AI is a tool. The only way students will get comfortable and good with the tool is by using it.
Learning Outcome: Students construct multi-part prompts that specify desired content constraints and stylistic parameters, demonstrating foundational prompt engineering skills, and establishing their role as supervisors of AI output.
C. PROFICIENCY IN DOCTRINE AND ANALYSIS THROUGH DIALOGUE (ONGOING: SOCRATIC BOTS)
The most consistent feature of the course is the deployment of the Socratic Dialogue Bots, which serves as the centerpiece of our strategy for enhancing doctrinal learning and analysis.
Six times during the semester, students engage in a private, ten-to-fifteen-minute Socratic dialogue with a specialized chatbot trained on the week’s reading materials and lecture notes.97
The students engage in these dialogues at the beginning of each unit, after doing all the class preparation but before engaging in any in-class lecture, discussion, or activities. This infrastructure intentionally scales the Socratic Method to the entire class, providing every student with structured, low-stakes practice that dramatically increases their opportunity to practice thinking on their feet—the analytical skill that the Socratic Method is designed to teach.98 Students upload the complete conversation transcript for review, providing me with continuous data on patterns of confusion and analytical miscues, which I can clarify in subsequent classes.99 I can also refer students who are struggling with case reading and analysis for additional academic support.
This approach directly addresses the concerns raised earlier about the Socratic Method as a “place of judgment, not curiosity” for neurodivergent and other students. By removing the peer audience and the punitive context of cold-calling, the bots eliminate the need for masking, resulting in a psychologically safe space for self-assessment and encouraging students to work through their confusion and learn without added emotional labor. One neurodivergent law
97 I needed to make sure that the six Socratic Bots replicated the Socratic Method, asked appropriate questions, followed up on student answers appropriately, and did not hallucinate. I asked several of my former high-performing Contracts students to “test” each Bot three times, taking on the persona of different types of students. They had mostly positive feedback of how the Bots performed and were impressed with them as teaching tools. However, Amanda Soderlind and I took all their critical feedback on each Bot to “tinker” with it to improve its performance and accuracy. I thank MHSL students Brittany Shields, Nichole Valdez, Sean Vinsel, Maja Watson, and Tara Westerlund for testing the Bots.
98 See Lande, supra note 48, at 6–7 (advocating for “Using Bots as Simulated Conversation Partners”).
99 See generally Harrington, supra note 2.
student who participated in testing the Socratic Bots confirmed this design benefit, reporting that the bot dialogues provided “a more enjoyable way for me to take in material” and noting the contrast with traditional classroom interaction, which creates “exhaustion and stress from the volume of interaction.”100 The instantaneous feedback loop provided by the Socratic Bots is proven to be a superior driver of long-term doctrinal retention.
Learning Outcome: Students demonstrate comprehension of case holdings and doctrinal rules through sustained dialogue, identifying gaps in their understanding through iterative questioning, and receiving immediate formative feedback in a psychologically safe environment.
D. LEARNING BY TEACHING (WEEK 3: BOT BUILDING)
Following the first Socratic Bot exercise, the next assignment requires students to engage with the doctrine at the highest level of mastery: teaching it.
In this low-stakes exercise, students build their own simple AI chatbot and train it to teach law students a foundational yet nuanced concept: consideration and its exceptions. To successfully build a teaching bot, students move beyond mere comprehension. They must structure the information logically, anticipate potential follow-up questions, and articulate the legal rules with precision. This task is rooted in the powerful learning science principle that explaining a concept deepens understanding, placing the burden of cognitive organization squarely on the student, thereby resulting in stronger doctrinal mastery.101 Educational research confirms that the protégé effect—learning by teaching—produces superior retention and understanding compared to passive study methods.102
Learning Outcome: Students demonstrate mastery-level understanding of foundational contract doctrines by designing and training an AI chatbot to teach those concepts, requiring them to structure legal knowledge logically, anticipate user confusion, and articulate rules with precision.
E. THE HUMAN EDITOR (WEEK 6: AI-ASSISTED BAR EXAM ESSAY)
This bar-exam-essay assignment is a core component of my strategy to address the “students won’t learn” objection by ensuring that human judgment remains paramount, even on timed, bar-exam-style assessments. The exercise has students work on a closed-book Contracts bar-exam-style essay in three distinct phases.
100 Anonymous student feedback from bot testing (Fall 2025) (on file with author); See also Kelly, Be Curious, supra note 19, at 248.
101 See Kelly, Be Curious, supra note 19, at 295.
102 See John A. Bargh & Yaacov Schul, On the Cognitive Benefits of Teaching, 72 J. EDUC. PSYCH. 593, 597 (1980) (finding that preparing to teach produces superior retention compared to studying alone); see also Kemp, supra note 49, at 638 (advocating for process-focused assignments that test collaboration and critical thinking).
- AI-Draft and Prompt Submission: The student first develops a sophisticated Setup Block (prompt) and feeds AI the provided Fence Fact Pattern,103 instructing the machine to produce a complete IRAC-style draft. The student submits their exact prompt transcript.
- Human Revision: The student then reviews the AI-generated draft, using Track Changes to revise and correct errors in fact and law, analytical shortcomings, and improper conclusions. This forces the student to demonstrate that they are actively supervising the AI and asserting their professional competence.
- Timed Essay: Finally, the student performs a fresh, sixty-minute, closed-book timed essay (targeting 800–1,100 words) of the Fence Fact Pattern. This replicates the pressure of the bar exam while compelling the student to generate the final product using only their own retained doctrinal knowledge and organizational skills developed during the revision process. This also ensures students have opportunities to practice writing clearly and logically.
This three-phase structure maps directly onto Bloom’s Taxonomy: the AI handles lower-order skills (articulating and applying rules), while the student focuses on higher-order cognition (analyzing the AI’s errors, evaluating conclusions, and creating an improved work product).104 This allocation allows me to assess what truly matters: the students’ analytical and synthesis skills. Students will still need to memorize black-letter law for the bar exam, which they cannot take with AI assistance. However, this technique allows them to practice bar-exam essay-writing during law school—either by generating an AI draft first and then critiquing it, or by writing their own draft first and using AI to identify weaknesses in their analysis. The goal is to expose students to different approaches to personalizing their learning, recognizing that individual students may find different learning methods more effective. Of course, whether AI-assisted practice improves bar exam performance remains an empirical question requiring further research.105
Instead of assessing the learning on the AI’s initial draft, the assessment focuses on the quality of the student’s revision and their performance on the unassisted timed attempt. This process directly trains the essential HITL meta-skill needed for modern practice.106
Learning Outcome: Students will be able to (a) generate sophisticated prompts for doctrinal legal analysis, (b) critically evaluate and revise AI-generated legal writing for accuracy, organization, and analysis, and (c) independently produce a timed legal essay demonstrating retained doctrinal knowledge and organizational and analytical skills developed through the revision process.
103 This is a previous California Bar Exam Contracts question that students had sixty minutes to write when it was administered. It was the third essay question on the July 2008 bar exam.
104 See generally ANDERSON & KRATHWOHL, supra note 21.
105 To my knowledge, no published studies have yet examined whether AI-assisted legal writing practice during law school correlates with improved bar exam essay performance. This represents a critical area for future educational research.
106 See Kemp, supra note 49, at 633, 642 (advocating for process-focused assignments that test collaboration and critical thinking); see Crootof et al., supra note 84, at 440–42 (on Human-in-the-Loop model); see also Stryker, supra note 85 (on Human-in-the-Loop oversight).
F. VERIFICATION AND ETHICAL AND HUMAN JUDGMENT (WEEK 7: JUDICIAL OPINION)
This assignment builds on the prior assignment and directly tests the limits of AI, addressing the ethical imperative of verification and compelling students to take responsibility for accurate research and fine-tuning their own critical judgment to counteract AI’s propensity to hallucinate. This assignment was developed in partnership with my law school classmate Judge Robert McBurney of the Superior Court, Fulton County, Georgia, and MHSL research and instructional librarians Alisha Hennen, Sonya Huesman, and Lisa Heidenreich, who contributed their expertise in legal research methodology and AI research tools to ensure the exercise authentically replicates how lawyers conduct research.
Students are presented with the complaint, answer, and contract in Hutter & Associates Inc. v. Hyde Brewing Co. and assigned a unique, non-Georgia jurisdiction. This case was heard before Judge McBurney, who livestreams his oral arguments.107 Students watch the actual oral argument on a motion for summary judgment in the case before drafting an opinion on the summary judgment motion under the law of their assigned jurisdiction.108 The case raises a foundational contracts formation question—whether a construction agreement constitutes a binding contract or merely an agreement to agree—but involves factually nuanced provisions that require careful application of the correct legal rules to specific facts. At the conclusion of the exercise, I will provide students with Judge McBurney’s actual written opinion, allowing them to compare their analysis with how an experienced jurist resolved the same question under the law of a different jurisdiction. Their task is to use legal AI tools (Lexis+AI, Westlaw Precision, Bloomberg Law) to research the case and draft a judicial opinion granting or denying summary judgment.
The objective is to expose students to the necessity of human oversight, but this goes far beyond simply checking for hallucinations. Students must ensure their research is correct, that the law stands for the propositions they claim, that they are engineering prompts to retrieve accurate legal authority, that the law is applied correctly to the facts, and that they are exercising human judgment to solve a complex legal problem correctly. They are required to verify every citation, ensure the law is current for their assigned state, and confirm that the cases say what they purport to say. They must submit a Prompt Management Tree documenting their research path, illustrating how they worked iteratively to manage the inevitable errors and biases inherent in generative AI legal research. Checking for hallucinations is a threshold competency; the real learning objective is developing the sophisticated verification and judgment skills that define competent lawyering. The stakes are high: if a student cites a hallucinated case, they fail the assignment.
107 See generally Complaint, Hutter & Assocs. Inc. v. Hyde Brewing Co., No. 2024CV005982 (Ga. Super. Ct. Fulton Cnty. Apr. 24, 2024); see generally Answer, Hutter & Assocs. Inc. v. Hyde Brewing Co., No. 2024CV005982 (Ga. Super. Ct. Fulton Cnty. Apr. 24, 2024). Judge McBurney’s practice of livestreaming oral arguments provides a unique pedagogical opportunity for students to observe the argument in real time before attempting their own analysis. Full video is available at https://www.youtube.com/watch?v=fu1eckGslrQ (last accessed Dec. 3, 2025).
108 See supra note 107.
Learning Outcome: Students will be able to (a) use AI legal research tools while engineering prompts to retrieve accurate legal authority, (b) verify all citations and confirm the law supports their claimed propositions, (c) apply legal rules correctly to factual circumstances through independent judgment, (d) document their iterative prompt refinement and research process, and (e) draft a judicial opinion demonstrating both technical competence in AI-assisted research and sophisticated analytical reasoning in resolving a complex legal question, thus transforming AI’s hallucination risk into a mandatory professional competency requiring human oversight.
F. PRACTICE-READY DRAFTING,REVISING, AND NEGOTIATING (SECOND HALF: SPELLBOOK SERIES)
The final series of exercises utilizes the specialized industry tool Spellbook to transition students into transactional practice, providing foundational exposure to the reality of an AI-augmented job market.109 This series includes scaffolded exercises across the second half of the semester that directly counter the concerns surrounding a professional succession gap. Rather than producing junior lawyers who are rendered redundant by AI-driven contract tools, we are striving to produce graduates who can use AI to enhance their transactional skills and accelerate their competency development while understanding the tools’ limitations. Despite legal organizations and law firms relying on AI and inevitably reducing junior lawyer hiring by a significant amount, firms and organizations will still need junior lawyers who can supervise these tools. This course provides critical early exposure to these competencies, though students will need sustained practice with AI tools across multiple courses throughout their legal education to develop the sophistication required for modern practice.110
The assignments are scaffolded:
- AI Review/Redline (Benchmark) (Week 8): Students use the Spellbook Benchmark feature to review a provided Lumon Non-Disclosure Agreement. The benchmark feature compares the provided draft against a proprietary database of market clauses, allowing the student to spot deviations and risks. This teaches issue identification by forcing the student to critique a machine-generated audit.
- AI-Assisted Negotiation (Negotiate) (Week 10): Next follows a live simulation where students, representing Lumon Inc., negotiate complex terms of a Master Services Agreement against a counterparty. Students use the Spellbook Negotiate feature to audit the opposing counsel’s contract for risk, compliance issues, and payment terms, utilizing the Ask feature for real-time strategic counsel.
- AI-Assisted Contract Drafting with Complex Facts (Week 12): Students draft a comprehensive employment agreement incorporating a detailed set of client requirements, demonstrating mastery in translating business objectives into enforceable contract terms through AI-assisted drafting. They edit the final AI product in Track Changes to demonstrate the limitations of AI in drafting and revising a contract.
- Summative Negotiation and Drafting Project (Week 14): In this capstone exercise, students work in pairs to negotiate a resolution to a commercial landlord–tenant dispute. Half the students represent the landlord, half represent the tenant. Students negotiate, draft, revise, and finalize an agreement that is their summative assessment for the second half of the course, demonstrating their ability to use AI tools strategically while understanding their benefits and limitations.
109 See Moppett, supra note 49, at 240, 257 (arguing for focusing assessment on the human revision process); Lande, supra note 48, at 6 (students can practice negotiation skills with bots).
110 See Lichtinger & Hosseini, supra note 54, at 4 (on junior employee headcount lowering 9 percent in organizations that adopt AI compared to non-adopting organizations).
Taken together, these exercises help students learn and refine their abilities to use legal AI tools effectively for contract drafting and revising, understand their limitations, and exercise the human judgment necessary to supervise AI-generated work. The main pedagogical focus is not simply teaching students to become proficient drafters since AI will increasingly handle that function in practice. Instead, students gain valuable experience in using AI as a necessary legal tool that requires sophisticated human oversight.
Learning Outcome: Students will be able to use industry-standard AI contracting tools to (a) identify risks and deviations in contract drafts through automated benchmarking, (b) draft and revise contract language with AI assistance while exercising independent professional judgment, (c) engineer prompts that generate provisions meeting client requirements, (d) evaluate and implement AI-suggested contract revisions, (e) conduct live negotiations using AI-generated strategic analysis, and (f) collaborate on resolving conflict, demonstrating mastery of both AI capabilities and limitations while recognizing that human supervision remains essential in AI-augmented practice.
***
Each assignment includes a detailed grading rubric shared with students at the time of the assignment, with assessment criteria that explicitly distinguish between the quality of students’ human work product and their effectiveness in using AI to improve that work product. For example, the Judicial Opinion assignment uses a three-tier structure: students achieve (a) Mastery, if they use sophisticated legal reasoning and complex problem-solving, as demonstrated through their independent analytical judgment layered on top of AI-verified research; (b) Proficiency, if they use AI correctly and verify all legal authority, ensuring the law and application are accurate; and (c) Needs Improvement, if their citations are wrong or their legal authority is inaccurate. This transparency in assessment criteria reinforces the trust-based learning environment and ensures students understand exactly what “competent AI supervision” means in practice—AI provides the foundational research support, but students must supply the critical thinking, contextual judgment, and legal reasoning that distinguish competent lawyering from mere information retrieval.
IV. ADDRESSING OBJECTIONS
The integration of AI into legal education’s core curriculum generates serious objections that deserve serious engagement: Will students pass the bar exam if they rely on AI during the semester? Will they develop the analytical skills that define competent lawyering? Will AI become a vehicle for academic dishonesty? Can this approach work beyond transactional courses? These concerns reflect educators’ and administrators’ genuine commitment to maintaining rigor and ensuring graduates are prepared for practice.111 But they rest on a flawed premise: that traditional legal education is pedagogically optimal and that AI is merely a shortcut to be policed.
I’ve heard every one of these objections—from colleagues, from administrators, from lawyers. They’re legitimate concerns that demand direct answers. This Part addresses each objection in turn, demonstrating how the course design transforms these concerns into learning opportunities. Part IV.A addresses bar preparation worries. Part IV.B tackles the concern that students won’t learn analytical skills. Part IV.C confronts the academic integrity question. Part IV.D responds to claims that this approach only works for Contracts.
A. “THIS WON’T PREPARE STUDENTS FOR THE BAR EXAM”
Bar passage rates matter. They matter to students, to schools, to employers, and to accreditors. The central concern is grounded in reality: the bar exam requires doctrinal knowledge and closed-book, time-constrained analysis. The assumption follows naturally: a course embedding AI encourages reliance on the machine, thereby eroding the fundamental memorization and recall necessary for bar passage.
1. No Evidence of Harm, Strong Evidence of Benefit
Here’s what we know: no empirical evidence (as of yet) demonstrates that AI-integrated pedagogy harms bar exam performance.112 And learning science strongly supports the opposite conclusion. The Contracts course design is built on retrieval practice and spaced repetition, methods that decades of cognitive science research have proven to be significantly more effective for long-term retention than traditional passive studying such as re-reading or listening to lectures.113 This approach therefore applies well-established science to legal education.114
Critically, students are learning how to build AI-powered tools that can help them prepare for the bar exam across all subjects, not just Contracts.115 The course teaches students to use AI to
111 See generally Moppett, supra note 49, at 272 (articulating concerns about maintaining analytical rigor in AI-integrated education).
112 It is still early, and research in this area is burgeoning. As empirical studies are published, we will understand more about AI’s impact on law students’ bar exam performance.
113 Henry L. Roediger, III & Jeffrey D. Karpicke, The Power of Testing Memory: Basic Research and Implications for Educational Practice, 1 PERSPS. PSYCH. SCI. 181, 185–89 (2006) (demonstrating that retrieval practice produces superior long-term retention compared to repeated studying); see also John Dunlosky et al., Improving Students’ Learning with Effective Learning Techniques: Promising Directions from Cognitive and Educational Psychology, 14 PSYCH. SCI. PUB. INTEREST 4, 35 (2013) (identifying practice testing and distributed practice as having “high utility” for learning).
114 See sources at supra note 112.
115 See AI and Neurodiverse Students, supra note 31 (demonstrating how students can use AI to generate practice hypotheticals, create customized outlines and study aids, receive iterative feedback on legal reasoning, and develop transferable learning strategies applicable across all bar-tested subjects).
generate practice hypotheticals, create customized study aids, receive feedback on their practice essays, and develop personalized learning pathways. These are transferable skills that apply to every bar-tested subject. This mirrors the pedagogical approach that Professor Niedwiecki describes as teaching students to create AI “study buddies” that function as always-available tutors capable of explaining concepts, generating practice problems, and providing patient feedback.116 A meta-analysis of feedback interventions confirms that immediate, specific feedback is among the most powerful drivers of student achievement.117 Rather than creating dependence, this approach builds metacognitive awareness about effective learning strategies and gives students technological competence they can deploy throughout bar preparation. The pedagogical principle is straightforward: use AI as a powerful learning tool during preparation but require students to demonstrate mastery without AI on bar-exam-style assessments— precisely replicating the conditions they will face on the actual bar exam (and in subsequent practice).
The Contracts course structure implements a two-pronged assessment model that compels preparation:
- Six Low-Stakes Quizzes: These frequent, formative assessments compel continuous engagement with doctrine. They are time-constrained and designed so that AI gives wrong answers, discouraging inappropriate reliance on generative text. Of course, the bigger consequence is that, if these quizzes are not taken seriously, students will not be prepared for the final exam. Crucially, the quizzes’ minimal impact on the final grade mitigates the high anxiety associated with testing and discourages AI use while maximizing the retrieval practice and spaced repetition that drive long-term retention.118
- Closed-Book Final Exam: The course culminates in a final exam that is explicitly closed-book, closed-internet, and multiple-choice, with questions previously used on the Multistate Bar Examination, precisely mirroring that format (a format that will inevitably need to be adjusted in the future as the NextGen Uniform Bar Examination becomes the standard).119 Students know from day one that the final, summative assessment requires internalized doctrinal knowledge. This transparency ensures that students focus their study on retention, regardless of the tools used during the semester assessment requires internalized doctrinal knowledge. This transparency ensures that students focus their study on retention, regardless of the tools used during the semester.
116 Professor Anthony Niedwiecki, conversations with author (Niedwicki is a fellow professor at MHSL, and he incorporated into his Business Organizations class a pedagogical approach where he created AI “study buddies” that function as personalized tutors for generating practice problems and providing patient, iterative feedback).
117 See John Hattie & Helen Timperley, The Power of Feedback, 77 REV. EDUC. RSCH. 81, 84 (2007) (meta-analysis finding feedback among the most powerful influences on achievement); see also Fleckenstein et al., supra note 46 (finding significant positive effects of automated feedback on student learning).
118 See generally Harrington, supra note 2 (discussing the benefits of low-stakes, frequent assessment in legal education).
119 See CORNELL UNIV. COMM. REPORT, supra note 75, at 34 app. G (recommending closed-book, in-person exams for foundational courses where AI-assisted work is permitted during the semester); About the Nextgen Bar Exam, NAT’L CONF. B. EXAMINERS, https://www.ncbex.org/exams/nextgen/about-next=gen (last visited Dec. 1, 2025) [hereinafter NCBE NextGen] (announcing first administration in July 2026 with phased adoption through July 2028).
This design works particularly well for in-person or synchronous courses where closed-book exams can be administered with traditional proctoring. Faculty teaching in fully asynchronous or blended formats without in-person exam administration face a legitimate tension: how to maintain assessment integrity without resorting to invasive remote proctoring. The answer lies in the same principle applied throughout the Contracts redesign: make AI use transparent and assess human value-added. For asynchronous courses, faculty might require process documentation (showing prompt iterations and revisions), implement oral exam components via video conference, design assessments that require synthesis across multiple sources where AI is less sophisticated, or accept that some risk exists while trusting that students who engage authentically with AI-enhanced learning will develop genuine competence that benefits them professionally regardless of exam format.
In the case of my in-person Contracts course, the redesign directly addresses the concern about bar preparation by replicating the conditions students will face on the actual bar exam. If we want students to have practice performing under bar-exam-like constraints, we must create those constraints: closed-book, time-limited, no AI access. The AI-enhanced work throughout the semester develops deeper understanding and stronger retention and analytical skills, but the closed-book final exam ensures students can retrieve and apply that knowledge without technological support. In many ways, AI takes the place of Examples and Explanations and other study aides that students have long used. This distinction between assessment contexts teaches students professional judgment to recognize which tools are required in which contexts.
Moreover, the bar exam format itself is currently in flux, which strengthens rather than undermines this pedagogical approach. Beginning in July 2026, many jurisdictions will transition to the NextGen Uniform Bar Examination, which represents a fundamental shift away from pure memorization toward skills-based assessment.120 As the National Conference of Bar Examiners explains, the NextGen exam is “designed to balance the skills and knowledge needed in litigation and transactional legal practice” and “will reflect many of the key changes that law schools are making today.”121 The exam emphasizes seven foundational lawyering skills—including legal research, legal writing, issue spotting and analysis, and client counseling—alongside reduced doctrinal coverage (eight subjects rather than fourteen).122 This shift validates exactly the kind of AI-enhanced pedagogy this course employs: using technology to develop deeper analytical skills and practical competencies rather than simply memorizing rules. The NextGen Bar Exam is “focusing less on memorization and more on skills,”123 which makes the learning processes students develop during an AI-integrated semester even more applicable to bar success. While actual NextGen practice questions are not yet widely available, the exam’s emphasis on applying knowledge in realistic contexts rather than recalling isolated rules aligns with AI-enhanced active learning.
Far from harming bar preparation, this AI-enhanced pedagogy may improve it. The course design provides students with significantly more practice opportunities and immediate feedback—like the bar-exam essay assignment—than a traditional lecture course. Most importantly, through the course assignments, students are learning how to make AI tools (along the lines of Professor Niedwiecki’s “study buddies”) that can help them prepare for the bar exam in all subjects.124
2. The Burden of Proof
The impact of AI-integrated legal education on bar exam performance compared to traditional pedagogy remains an empirical question that warrants formal study. Critically, the bar exam in general is currently in flux given the current transition to the NextGen Bar Exam.125 Owing to the lack of empirical data around the impact of AI-integrated legal education on law student performance, I will be conducting a pilot study to explore the effectiveness of the course redesign in spring 2026. But absent any evidence of harm, the burden of proof should rest on those claiming that established pedagogy practices—like retrieval practice, spaced repetition, and immediate feedback—when augmented by AI, will somehow fail in legal education when they have succeeded across other disciplines.126
123 NextGen Bar Exam: About the Exam Changes in 2026, KAPLAN, https://www.kaptest.com/study/bar/new-next-gen-bar-exam/ (last accessed Dec. 1, 2025) (“NextGen UBE’s goal is to provide a practical exam that focuses more on testing examinees’ skills and abilities than the UBE does. While testing legal knowledge is still a focus, the recalibration is a direct response to concerns and research that showed the UBE focused too heavily on pure memorization.”).
124 See supra note 115.
125 See NCBE NextGen, supra note 118.
126 See PETER C. BROWN ET AL., MAKE IT STICK: THE SCIENCE OF SUCCESSFUL LEARNING 3–28 (2014) (documenting decades of research establishing that retrieval practice, spaced repetition, and immediate feedback enhance long-term retention across disciplines); Alice Latimier et al., A Meta-Analytic Review of the Benefit of Spacing Out Retrieval Practice Episodes on Retention, 33 EDUC. PSYCH. REV. 959, 980 (2021) (meta-analysis confirming robustness of spacing effect); Campbell R. Bego et al., Single-Paper Meta Analyses of the Effects of Spaced Retrieval Practice in Nine Introductory STEM Courses, 11 INT’L J. STEM EDUC. 1, 15 (2024) (demonstrating effectiveness in undergraduate STEM education); Rachel Gurvich et al., Reimagining Langdell’s Legacy: Puncturing the Equilibrium in Law School Pedagogy, 101 N.C. L. REV. 118, 118 (2023) (arguing that traditional legal education was “devised for a different set of students and at a time when we knew far less about how people learn”).
This gets to something bigger: pedagogy has continually evolved over time as society and technology have changed: incorporation of skill-based and practical education, the emergence of hybrid and online teaching, the increasing importance of critical theory. There has never been empirical data to verify that a pedagogical method is sound and effective before pioneering educators innovate, develop new approaches, and test them with students in the classroom.127 We are currently in the innovation and experimentation phase of AI integration into legal pedagogy. If we’re going to reject AI integration based on bar exam concerns, we should first ask whether current pedagogy is actually maximizing bar success. Or is it simply maintaining the status quo?
B. “STUDENTS WON’T LEARN LEGAL ANALYSIS”
The worry is straightforward: by delegating routine tasks to the machine, students will fail to develop core analytical competencies and never learn to “think like a lawyer.” If AI writes the first draft, the concern goes, students will not develop the cognitive muscles that legal reasoning requires.
I have also heard the unspoken version of this objection, which seems to be rooted in the previously mentioned outdated academic hierarchy that assumes legal writing and research professors—who have been leading the way in integrating AI into their teaching—are teaching “skills” rather than “real” doctrine or law.128 This assumption is wrong on two levels.
1. The Skills and Doctrine Distinction Is Artificial
Reading cases, developing rules, applying rules to facts, and constructing and evaluating arguments are critical thinking skills that we must help law students develop.129 Professors teaching doctrinal courses have always taught students critical thinking skills: how to think like lawyers. Success in the AI era requires lawyers to possess robust critical thinking abilities to evaluate AI output and handle complex work that AI cannot replicate.130 AI can be a powerful tool for practicing these very skills. The course design provides students exponentially more practice with doctrinal application than is possible in a Socratic or lecture-and-discussion class, precisely because AI handles the production while students focus on judgment.
2. More Practice, Not Less
Exercise in the course compels the development of core analytical skills. For example:
127 Still, legal pedagogy has been slow to evolve. See generally WILLIAM M. SULLIVAN ET AL., EDUCATING LAWYERS: PREPARATION FOR THE PROFESSION OF LAW (2007) (criticizing the stasis in legal pedagogy and calling for evidence-based reform).
128 See Kemp, supra note 49, at 638 (discussing the perception that AI threatens “real” legal education).
129 See Kelly, Be Curious, supra note 19, at 295–96 (arguing that legal reasoning is a skill that can be taught through multiple modalities).
130 Moppett, supra note 49, at 257 (arguing that “as AI assumes a more prominent role, the importance of critical thinking skills among legal professionals becomes even more pronounced”).
The Bot Building Exercise requires students to achieve mastery-level understanding of doctrine. To train a bot to teach consideration, a student must structure the information logically, anticipate follow-up questions, and articulate rules with precision. The crucial cognitive benefit is that AI cannot fake this understanding; the student must possess it, a process proven to produce superior retention compared to studying alone.97
The Socratic Dialogue Bots provide structured, low-stakes practice sessions that dramatically increase students’ opportunity to practice analytical thinking. Six times during the semester, students engage in private, ten-to-fifteen-minute Socratic dialogues with specialized chatbots trained on the week’s reading materials and lecture notes. By scaling the Socratic Method to the entire class and removing the peer audience, the bots ensure every student gets repeated practice thinking on their feet—the analytical skill the Socratic Method is designed to teach.98 Students upload complete conversation transcripts for review, providing continuous data on patterns of confusion and analytical miscues that can be clarified in subsequent classes.
The AI-Assisted Bar Exam Essay requires students to revise AI-generated analysis using Track Changes, demonstrating their supervisory role over the machine. Students must identify logical failures, correct misstatements of law, and improve analytical depth. The assessment measures the human revision, not the AI draft. The final step is the most critical: students must write a timed bar exam essay, unassisted by technology—testing their ability to produce coherent legal analysis without input from AI.
The Judicial Opinion Assignment gives students extensive practice in verification and critical judgment by requiring them to work with AI-generated legal research that inevitably contains errors. Students must verify every citation, check that cases say what they claim, and ensure legal accuracy—practicing the iterative process of catching and correcting AI mistakes. The three-tier grading rubric reinforces that verification alone isn’t enough: students must practice layering sophisticated legal reasoning and analytical judgment on top of verified research, distinguishing between AI’s role in retrieving information and students’ crucial roles in validating research and applying critical thinking and contextual understanding to create sound legal analysis.
The entire course design is predicated on the principle that AI is an enhancer—as my friend Alex Arbit called it—instead of a replacement.99 Students learn both doctrine and critical evaluation of AI because assignments require human judgment at every stage.
C. “THIS WILL CAUSE STUDENTS TO CHEAT”
This objection is often the most emotionally charged, highlighting how AI is causing academic dishonesty and preventing that dishonesty should be our focus as educators. Students are using AI to complete assignments without learning, and institutional integrity is being compromised, they assert. When I first proposed this course, this was the concern I heard most often. And I
131 See Bargh & Schul, supra note 101 (finding that preparing to teach produces superior retention compared to studying alone).
132 See Porter, supra note 33.
133 Interview with Alex Arbit, Software Engineer (Nov. 2025).
understand where it comes from. We all have heard about students using AI when it has been prohibited. We all have suspected AI use in student work. The fear is legitimate.
But the prevailing institutional response of deploying remote proctoring software and AI detection tools is both ineffective and destructive to learning; with detection tools both inaccurate and biased, and monitoring tools posing serious privacy and equity concerns.134 Moreover, surveillance-based approaches create a “low-trust environment” that damages the student-faculty relationship essential to learning.135 A focus on surveillance also takes faculty time and energy away from high-value tasks.
1. Design for Trust, Not Surveillance
As discussed earlier, the Contracts course rejects the policing mindset entirely. The pedagogical question is “How do I design assignments where AI enhances learning?” The course is built on transparency and trust. AI use is required and assessed instead of being merely permitted (or, to the opposite extreme, forbidden). Students know exactly what is expected: use the tools, but take responsibility for the output.
The assignment design makes cheating either impossible or self-defeating:
- Prompt Submission Requirements: Students submit their exact AI prompts alongside their work, making the collaboration transparent and assessable.
- Track Changes Requirements: Students must demonstrate their improvements over AI output, proving they engaged critically with the machine’s work. For the Spellbook assignments, they will also be required to write a memo to a supervising attorney explaining why they made the changes they made, further engaging them in thinking critically.
- Verification Requirements: In the Judicial Opinion assignment, students cannot outsource verification to AI; they must do it themselves and show their revisions in Track Changes.
- Closed-Book Assessments: The timed final exam requires demonstrated individual competence without AI assistance.
2. The Trust Dividend
134 See Elkhatat et al., supra note 57, at 17 (reporting high inconsistency and false positive rates); Dalalah & Dalalah, supra note 57; see also Balash, supra note 60, at 263; see also Mita, supra note 62, at 1553.
135 See Luo, supra note 59, at 991, 998 (finding that detection-focused approaches create “low- trust environment[s]” where students feel “unsafe to freely explore GenAI use”).
This approach aligns with emerging research suggesting that clear, transparent AI policies produce better learning outcomes and reduce anxiety about appropriate use rather than surveillance.136 More fundamentally, it treats students as future professionals.137
If we make AI use transparent rather than hidden, when we assess the human value added rather than the raw output—students will rise to meet that standard. The course design assumes professional responsibility instead of policing for dishonesty. And that assumption, in my experience, tends to be self-fulfilling.
D. “THIS APPROACH ONLY WORKS FOR CONTRACTS”
A final objection holds that AI integration is uniquely suited to Contracts because of its transactional focus and would fail in litigation-oriented courses like Torts, Civil Procedure, or Constitutional Law. This sells both AI and the pedagogy short.
1. Doctrine-Agnostic Pedagogy
The core pedagogical architecture—retrieval practice, immediate feedback, required verification of AI output, and assessment of human value-added—is entirely doctrine-agnostic.
The Socratic Dialogue Bots work for any doctrine. A bot can be trained on negligence elements (Torts), personal jurisdiction (Civil Procedure), or First Amendment standards (Constitutional Law) as readily as on contract law. The pedagogical benefit of scaled, individualized Socratic dialogue with immediate feedback is identical regardless of subject matter.
The Bar Exam Practice Model applies to any bar-tested subject. The three-phase structure (AI draft → human revision → independent timed essay) works for a Torts hypothetical, a Civil Procedure fact pattern, or a Constitutional Law issue-spotter.
The Human Editor Approach transfers directly. Having students critique and improve AIgenerated legal analysis develops critical thinking skills regardless of the underlying doctrine.
What changes across courses is the industry tool instead of the pedagogy. Contracts uses Spellbook for drafting and negotiation; a Litigation course might use AI-assisted brief analysis or discovery review tools; a Constitutional Law course might use AI for legislative history research or policy analysis. The tool serves the pedagogy, and different tools can serve the same pedagogical goals.
2. Contracts as a Hard Case, Not Easy Case
136 Bjarne D. Lund et al., AI and Academic Integrity: Exploring Student Perceptions and Implications for Higher Education, 23 INT’L J. EDUC. INTEGRITY 1545, 1561 (2025) (finding that clear institutional policies reduce student anxiety about AI use); Tight, supra note 67, at 920 (noting “surprisingly little evidence” that surveillance-based approaches reduce cheating).
137 See sources cited at supra note 135.
The potential success of this model in a required, bar-tested, first-year course demonstrates its applicability to any foundational course in legal education. Contracts is not an easy case for AI integration. It is a hard one. Students must master consideration, offer and acceptance, statute of frauds, and remedies to pass the bar. If AI can be embedded in Contracts without sacrificing rigor, it can be embedded in any doctrinal course.
The broader implication is significant: if this course design is effective, it may serve as a template for other courses. Faculty in other doctrinal areas could adapt the pedagogical framework while selecting AI tools appropriate to their subject matter.138 This begs the question whether we as legal educators are willing to rethink legal pedagogy across the curriculum.
V. CONCLUSION: REDEFINING THE PROFESSOR’S ROLE
A. EMPOWERED, NOT REPLACED
The journey of redesigning this Contracts course began with a personal and existential question: What do I offer that a bot does not? The initial fear that generative AI could render me redundant was fueled by the realization that an AI Professor could easily perform the traditional role of a human professor.139 AI can generate lectures, answer student questions, give hypotheticals and feedback, and even administer assessments.
The concern that AI-embedded pedagogy eliminates the need for faculty reflects a misunderstanding of both AI’s capabilities and the faculty’s essential role. Research confirms that while AI-generated instructional materials can achieve equivalent learning outcomes to human-created materials, students consistently prefer human instruction in terms of learning experience.140 This distinction between learning outcomes and learning experience is precisely why I do not believe AI should, or will, replace faculty. Instead, AI empowers the professor by liberating them from synchronous content delivery, cold-calling, and engaging students in realtime discussion and dialogue—and gives them back the time to be fine-tuned into each student’s learning. The core competencies of lawyering that AI cannot replicate remain firmly in the human domain: AI cannot provide empathy, exercise ethical and professional judgment, or engage in complex problem-solving and negotiation strategy.141 Therefore, the professor’s role must shift from that of a deliverer or questioner of content to a designer of learning experiences, facilitator of student engagement, and guide in developing junior lawyers prepared for an AI-augmented profession.
B. FROM TEACHER TO FACILITATOR
138 See Danault, supra note 1, at 1, 6 (discussing how faculty across disciplines can adapt AI tools to their pedagogical goals while maintaining their role as “facilitators of the learning process”).
139 Ladany, supra note 3.
140 Netland et al., supra note 4.
141 See Kemp, supra note 49 (identifying empathy, judgment, and strategic thinking as core competencies AI cannot replicate); see generally Moppett, supra note 49 (arguing that human judgment becomes more important as AI assumes routine tasks).
Instead of aiming to automate the professor’s job, the goal of embedding AI is to automate the content delivery, discussions, questioning, and low-stakes assessment that previously consumed my energy. The new, AI-embedded model places the student at the center of a dynamically designed learning ecosystem. The professor’s evolved role shifts to that of the architect of learning experiences and the mentor for individual growth.142 The work now involves:
- Designing learning experiences and curating the AI tools and exercises;
- Monitoring individual student understanding through assignment and transcript review and intervening with targeted support;143
- Focusing on higher-order cognition including evaluation and problem-solving— which is where the human element is essential;144 and
- Modeling professional and ethical judgment.145
This is the professor’s liberation, allowing a move away from the performance demands of traditional teaching. For a neurodivergent professor, this approach eliminates the need for masking by focusing on course design and one-on-one connection.146 The new role allows me to focus energy on what I do best, facilitating student-centered learning in the truest sense.
C. THE INSTITUTIONAL PERSPECTIVE: SLOW ROLLOUT AND MISSED OPPORTUNITY
Despite the profound pedagogical opportunities, the integration of AI into legal education has been characterized by a slow, cautious rollout. This hesitation is understandable, rooted in legitimate concerns about academic integrity and the potential for professional displacement.147 Indeed, only 11 percent of higher educational institutions have a comprehensive AI strategy, and a “whopping 74%” of CTOs view generative AI as a moderate or significant risk to academic integrity.148
142 See Danault, supra note 1, at 1, 5 (describing faculty role evolution from “content deliver[ers]” to “facilitators of the learning process”).
143 Harrington, supra note 2 (describing how bot transcripts enable individualized intervention and support).
144 Zhang et al., supra note 2, at 1–2 (finding that AI-enhanced pedagogy improves higher-order cognitive skills when properly implemented); see also Moppett, supra note 49, at 256–59 (discussing how AI enables focus on analysis and evaluation rather than rote learning).
145 See John Haddock, Harvey’s Principles for AI Adoption and Rollout in Law Schools, HARVEY (Oct. 31, 2025), https://www.harvey.ai/blog/harveys principles-for-ai-adoption-and-rollout-in-law-schools (emphasizing the importance of “preserving the rigor, ethics, and human judgment at the heart of the profession” within legal education in the AI era).
146 Kelly, Be Curious, supra note 19, at 274–75 (discussing how alternative pedagogical approaches reduce masking demands for neurodivergent educators); Miller, supra note 20 (describing the cognitive burden of masking in traditional teaching environments).
147 See supra Part IV.C (discussing legitimate concerns about academic integrity and the surveillance-based institutional responses).
148 2025 CIO/CTO Survey: Navigating the Complexities of AI Integration in Higher Education, JENZABAR 6 (2025), https://www.jenzabar.com cio-cto-survey (reporting that only 11 percent of institutions have comprehensive AI strategies and 74 percent view AI as moderate to significant
risk).
However, this caution comes at a significant cost. Law schools risk falling further behind the legal profession—which is rapidly adopting AI tools—as well as incoming students themselves, who will increasingly arrive with undergraduate AI expertise and expectations that legal education will match their technological sophistication.149 Moreover, remaining passive and prioritizing fear over exploration means institutions will effectively “cede intellectual leadership” to technology companies.150
The alternative is mission-driven adoption. In the case of MHSL and our access-driven mission, AI must be integrated first because it has the potential to create a more equitable learning environment for diverse learners, reducing resource gaps by democratizing personalized tutoring.151 Second, this integration is critical because it fulfills the school’s duty to prepare practice-ready professionals by training them in the essential meta-skills of AI literacy and verification.152
D. THE ROAD AHEAD
This Article has provided the theoretical framework and the practical implementation—the “how” of an intentional, mission-based design for what may very likely be one of the first AI embedded required doctrinal courses in legal education. But whether AI-integrated pedagogy produces equal or superior performance compared to traditional methods remains an empirical question.
To test this hypothesis, I will be conducting a pilot research study simultaneously with the Spring 2026 Contracts course. The study will examine learning outcomes, student experiences with AI tools, and equity impacts across different student populations.
I am committed to transparently reporting all results—successes and failures—in a follow-up Article. If the bot-building exercise flops, I’ll report it, plus specific lessons learned. If students struggle with a certain assignment, I’ll report it. The goal is to provide empirical data that can inform the ongoing conversation about AI in legal education.
149 See supra Part I.B (discussing rapid AI adoption in legal practice and the widening gap between legal education and professional practice); see also Natasha Singer, College Students Flock to a New Major: A.I., N.Y. TIMES (Dec. 1, 2025), https://www.nytimes.com/2025/12/01/technology/college-computer-science-ai-boom.html (reporting that MIT’s new “AI and decision-making” major is now the second-most-popular undergraduate major with nearly 330 students enrolled, and that the University of South Florida enrolled over 3,000 students in a new college of artificial intelligence and cybersecurity this semester, as dozens of universities create specialized AI programs and traditional computer science enrollment declines).
150 Underwood, supra note 91 (warning that institutional passivity will cede educational leadership to technology companies).
151 See supra Part I.C (discussing how AI democratizes access to personalized tutoring and reduces resource gaps); see also Fleckenstein et al., supra note 46 (finding AI-enhanced feedback particularly beneficial for under-resourced students).
152 See supra Part III (describing exercises that develop AI literacy and verification skills as essential professional competencies).
We are still at the beginning of understanding AI’s transformative impact on legal education. We don’t yet know what we don’t know about optimal AI integration, which makes transparent research and honest reporting of both successes and failures essential.
This is an invitation to colleagues across the country to experiment with this pedagogical framework, adapt it to their own contexts, and share their findings. Legal education needs a body of evidence, not just theoretical arguments. We need multiple faculty, at multiple institutions, trying different approaches and reporting what actually happens in the classroom.
E. THE ULTIMATE GOAL: ACCESS TO JUSTICE THROUGH ACCESS TO EDUCATION
The ultimate goal of this project is inextricably connected to MHSL’s vision of a “legal system that is just and accessible to all.”153 AI is a pathway to access. AI makes legal education more accessible to diverse learners, removing emotional and cognitive barriers that the traditional classroom creates and offering more diverse and dynamic learning modalities beyond lecture based and discussion-based courses.154
Future lawyers, having trained with these tools, will be equipped to leverage AI to make legal services more efficient and affordable for their clients, thereby improving access to justice.155 If we train students to use AI responsibly in law school, they can use it to expand their practices to serve clients who currently cannot afford legal representation. They might build AI-powered tools to help pro se litigants navigate family law matters or housing disputes. They could enable social justice and legal services organizations to take on more cases by automating routine document review and drafting, allowing limited resources to stretch further and serve more people who need representation—helping expand the circle of community members who can access legal help.
This project echoes Mitchell Hamline’s pioneering spirit: a full circle from the mission-driven design of the hybrid JD program, which was initially derided as “The Worst Idea Ever” but ultimately ushered in a new era of online learning in legal education.156 Because mission-driven innovation, even when initially resisted, can transform legal education for the better.
153 Mission Statement, MITCHELL HAMLINE SCH. OF LAW, https://mitchellhamline.edu/about/mission-vision-values/ (last visited Nov. 22, 2025).
154 See supra Part I.B (discussing Universal Design for Learning principles and accessibility benefits of AI integration); see generally Rose & Meyer, supra note 40 (foundational text on UDL principles).
155 See Absher, supra note 81 (discussing how AI tools enable lawyers to serve clients more efficiently and affordably); Arredondo, supra note 6 (“So, AI-powered technology such as LLMs can help to close the access to justice gap? Absolutely. In fact, this might be the most important thing LLMs do in the field of law. The first rule of the Federal Rules of Civil Procedure exhorts the ‘just, speedy and inexpensive’ resolution of matters. But if you asked most people what three words come to mind when they think about the legal system, ‘speedy’ and ‘inexpensive’ are unlikely to be among the most common responses. By making attorneys much more efficient, LLMs can help attorneys increase access to justice by empowering them to serve more clients.”).
F. A PERSONAL NOTE
For me, the AI transformation has been a profound personal shift. My twenty-four-year teaching career had become defined by teaching as an exhausting performance, from preparing lectures, to delivering content, and constantly managing the anxiety of being observed and evaluated in real time. It was draining in ways that went beyond normal professional fatigue.
Designing the AI-embedded course, however, has been joyful because it is playful and new. The work of designing bot personalities, crafting prompts, and imagining new ways for students to engage with doctrine feels creative instead of performative. It feels like building something rather than delivering something.
This renewed sense of professional purpose is perhaps the most compelling argument for embracing this transformation. The course innovation is undoubtedly hard and uncertain. I don’t have all the answers. I don’t know if everything I’ve designed will work. But it is also energizing and meaningful.
This is an invitation to all my colleagues to look at the tools of the future not as a threat to your expertise, but as a path to reclaim the joy of creation in your own professional work. We know that AI is already transforming legal education. Let us shape that change intentionally, guided by our institutional missions and our values, to create more effective and equitable classrooms to educate future lawyers.
156 See generally Janus, supra note 13 (recounting initial resistance to Mitchell Hamline’s hybrid JD program and its ultimate success in expanding access).