A law professor with no coding background demonstrates how law school educators can build effective AI-powered practice tools for law students designed to enable critical thinking instead of replacing it — the key is to start with pedagogy, not technology
Key highlights:
-
-
-
Design before you build— A professor at the University of San Francisco School of Law maps out the student interaction and learning objective before touching any platform — working backwards from there toward the desired outcome.
-
Constraints protect the learning process — Every tool is deliberately engineered to withhold answers thereby forcing students to lead the process and do the thinking themselves.
-
Low-stakes practice, high-stakes results— Students who rehearse privately with AI tools arrive in class more confident and can point to those simulations as real skills in job interviews.
-
-
The image of a law student buried in case books still rings true, but at the University of San Francisco (USF) School of Law, there is a new kind of study partner in the room. AI-powered assistants are helping students practice the Socratic method in private, master Bluebook citations without a professor, and simulate client interviews before setting foot in a real law office.
These tools, constructed as carefully designed learning environments, have been built by Prof. Nicole Phillips, co-director of the legal research, writing, and analysis program at USF, who openly admits she is not a tech person.
Over the past two years, Prof. Phillips has built several AI-powered tools that include a case brief helper, a mediation bot and an employment law counseling coach. Using platforms accessible to anyone willing to think carefully about pedagogy, she outlined the replicable steps for other law schools to follow, including:
Step 1: Start with a problem — The single biggest mistake faculty makes is starting with technology. “It can’t just be ‘Let’s use AI!’ There has to be a specific learning outcome,” Prof. Phillips says, adding that every tool she has built began with a concrete student frustration or gap. For example, students wanted more ways to practice Bluebook citations with real feedback, so she built a Socratic method tool after observing that student anxiety about the technique was interfering with their ability to demonstrate what legal knowledge they knew.
Step 2: Design the interaction before the build — Once the problem is clear, Prof. Phillips says she maps out the student experience before touching any platform. “I’m really thinking about what I want the students to get out of it and then working backwards from there.” This means deciding whether the tool should help students explain a concept, revise a draft, or respond to follow-up questions under pressure. Crucially, this design-first approach also forces the builder to define constraints. In fact, none of Prof. Phillips’s tools will give a student the answer; instead, the student must lead, and the tool follows and pushes back.
Step 3: Build in the constraints — The most important step in the build process is to take the risk of AI providing answers and engineer it out of the tool entirely. The Bluebook Citation Bot, for instance, will never produce a complete citation on demand. Instead, the goal is for students to understand why a citation is constructed the way it is. Similarly, the Socratic Method assistant is designed so that students must drive their own thinking and sit with the same discomfort that arises in a real classroom, but in a private space in which the stakes are lower.
Step 4: Try to break the tool — Before any tool reaches a student, Prof. Phillips tests it exhaustively: first, by feeding it incorrect law to see if it pushes back; and then, by probing every way it might accidentally give away an answer. “I do a lot of testing and breaking and then rebuilding,” she explains.
Step 5: Pilot and iterate — When a tool is ready, Prof. Phillips tells students what it is designed to do, what she hopes they will get out of it, and that they may find errors. To address any tool’s mistakes, she invites students to bring the errors to her. This improves the tool through real-world feedback that no solo testing can replicate, and it repositions students as collaborators in the learning design rather than passive recipients of it.
Of all her tools, Prof. Phillips considers the Socratic Method assistant the most consequential. For first-generation law students especially, the Socratic classroom can feel less like a learning environment and more like a barrier. “Competence is often mistaken for confidence,” she says. “The opportunity to practice being wrong privately is really important.” Students who use the tool arrive in class more willing to participate. For those who use her experiential simulation tools, she describes how students can point to their experience with them in job interviews noting that they have practiced these skills.
However, the biggest barrier to faculty building their own tools is the mistaken belief that it requires technical expertise. Admittedly, the hard part that Prof. Phillips insists on is the design. Her advice to her peers, however, is to start with a problem your students have, work backwards from what you want them to be able to do, build in the constraints that protect the learning, and then, break it before they do.
Learn more about the AI and Future of Legal Practice initiative here
