One of the challenges of course design is that all our work can seem perfectly clear and effective when we are knee-deep in the design process, but everything somehow falls apart when deployed in the wild. From simple misunderstandings to complex misconceptions, these issues typically don’t reveal themselves until we see actual student work—often when it’s too late to prevent frustration. While there’s no substitute for real-world testing, I began wondering if AI could help with this iterative refinement. I didn’t want AI to refine or tweak my prompts. I wanted to see if I could task AI with modelling hundreds of student responses to my prompts in the hope that this process might yield the kind of insight I was too close to see.
