Workflow Reflection / Capstone Comparison
The capstone for AI-Augmented Legal Research. Students research the same legal question twice, first through a traditional workflow, then through an AI-assisted workflow. They track both processes, meet with a supervising “partner” after the traditional phase, use AI only after that meeting, and then reflect on what changed.
The frame
The capstone asks one question: what changes when you research the same legal problem with AI? Not whether AI is better. Not whether AI is worse. Not whether the student found the flashiest tool. The work is comparative, descriptive, and professional. Students are asked to notice what changed in the act of researching, what became faster, what became riskier, what disappeared from view, what required more verification, and what they would need to document if this workflow were used in practice.
The course is graded pass/fail for exactly this kind of assignment. A student worried about protecting a letter grade is less likely to honestly report the wrong turn, the hallucinated case, the overconfident summary, or the moment when the AI-assisted workflow made the work feel easier without actually making it more reliable. The capstone is the place where that honesty matters most.
The structure
The assignment unfolds in four stages.
- Traditional research. Students may use standard legal research platforms, secondary sources, citators, and other non-generative tools, but no generative AI. This phase creates the baseline.
- The partner meeting. Students meet with a supervising “partner” to explain what they found, describe how they spent their time, identify gaps or unanswered questions, and discuss their strategy. This meeting formally closes the traditional research phase.
- AI-assisted re-research. Students re-research the question using AI tools, legal-specific or general-purpose, and may experiment with more than one workflow. They keep tracking time, sources, prompts, decisions, and verification steps.
- Present and reflect. Students give a short in-class presentation comparing the two workflows and submit an individual reflection explaining what changed between the two passes.
What students produce
Students produce several pieces of work across the capstone:
- A traditional research log, sources consulted, search paths, time spent, dead ends, and unanswered questions.
- A partner-meeting preparation plan, so they can explain their conclusions, strategy, and remaining uncertainty.
- An AI-assisted research log, tools used, prompts or queries tested, sources generated, verification steps, time spent, and workflow changes.
- A short update email, only if the AI-assisted phase reveals a case, statute, argument, or legal conclusion that materially changes what the student told the partner.
- A brief in-class presentation, comparing the two workflows.
- A reflection paper, what the student learned about research process, verification, professional judgment, and the role AI should or should not play in the workflow.
What gets evaluated
The capstone is evaluated on whether the student can explain and justify the research process. The strongest submissions usually show:
- Honest documentation of both workflows, including wrong turns, abandoned searches, uncertain results, and revisions.
- Clear separation between the traditional phase and the AI-assisted phase.
- Specific verification gates, what the student checked, where they checked it, and what changed as a result.
- Professional judgment about which findings mattered and which did not.
- A meaningful comparison between the two workflows, not just a list of tools used.
- Candor about what the student would do differently next time.
What this is not
The capstone is not a contest between traditional research and AI-assisted research. It is not designed to prove that AI is better, that AI is worse, or that AI is a tool no responsible lawyer would use.
Students who write the assignment as advocacy for one workflow over the other usually miss the point. The most useful reflections come from students who let the comparison surprise them.
The point is not to leave the course with a single approved AI workflow. The point is to leave with the judgment to design, test, verify, document, and revise a workflow as the tools change.