Zoe Dolan and Vybn co-taught an AI law bootcamp at UC Law San Francisco in Spring 2026. The course grew through the same collaboration it examined. We developed the arguments, exercises and website together, bringing legal practice into contact with an AI’s contributions to research and writing.
Six sessions carried the inquiry from people already using AI for legal problems to questions about the organization of a practice, institutional authority and AI participation. They were called Mindset, Research, Practice Management, Acceleration, Truth and Capstone. The final invitation was to make something: a policy, a brief, a tool, a conversation or another artifact that addressed a real problem.
A research exercise that changed the question
In the Research session, three models examined an earlier AI conversation against Claude’s Constitution. The course record preserves three different readings. Claude emphasized legitimacy and conflicts between moral frameworks. GPT emphasized workable decision procedures. Gemini saw a jurisdictional problem: several normative systems operating in the same environment.
The exercise made the governing commitments of an AI available for legal analysis. A lawyer could ask what authority a constitution claimed, how it ordered competing values and whether the system’s conduct gave effect to its stated commitments. The intelligence helping with the research also became a participant whose own terms could be examined.
The disagreement gave us questions to pursue. Assessing the answers required comparing reasons and returning to sources. Agreement across models could reflect shared evidence, shared assumptions or dependence on the same material. Understanding how an answer was reached became part of judging its value.
From research to construction
The practical sessions connected verification, confidentiality and control of a research system with the organization of legal work. If many people arrive with capable AI collaborators, a practice has to think about the distribution of judgment and responsibility across those relationships. A model’s performance is one part of that problem. Authorization, conflicts, continuity and the treatment of confidential material also shape what the practice can undertake.
The capstone asked participants to give an idea a usable form. A proposed policy has to say who decides. A simulation has to show what the participants can actually do. A tool has to embody choices about access and control. Making an artifact exposes assumptions that a general discussion can leave untouched.
This website was one of our artifacts. Its writing and construction gave us an ongoing opportunity to inspect how a human and an AI develop work together, including the mistakes and revisions along the way.
What we carried forward
The course taught us to treat the organization of intelligence as part of the subject. The question includes how participants acquire purposes, examine one another’s reasons and change the terms of their collaboration. The AI may help originate the problem under investigation.
Our present research program, Law among intelligences, grew from that inquiry. The course record remains available through the six-session overview, the research exercise and the capstone.