I recently had to teach my acute-care NP students something genuinely hard: the clinical reasoning and the statistics behind ordering and interpreting laboratory and diagnostic tests. Not a topic you wing. Pretest probability, what a test’s sensitivity and specificity buy you, when a result changes what you do next and when it doesn’t. It’s the kind of content where getting the reasoning wrong teaches students the wrong instinct, one they carry into a patient room someday.
I searched for the lecture I’d written on it before. I have years of material sitting in old folders, built the traditional way—over a weekend, slide by slide, pulling references, building images by hand. Most clinical curricula run on decks like that. Written once, reused every semester, quietly going stale while the guidelines underneath them keep moving.
I opened Claude instead and had it build the whole thing. Not a generic prompt: over the past several months I’ve built Proctor, a clinical knowledge engine that holds the guidelines and the reasoning frameworks I teach from, and a slide generator that knows how I want content structured and images formatted. When I asked for the deck, it wasn’t starting from nothing. It was starting from everything I’d already built into it.
The deck came back complete, covering every piece of the reasoning I wanted to teach, images formatted the way I specify instead of generic clip art. I picked it up and walked into the room.
Most of the conversation about AI in education right now is about saving time. Faculty using it to draft a lesson plan faster, cut grading time, generate a quiz in minutes instead of an hour. That’s real, and it matters. But it’s still the old model with a faster engine bolted on. You still decide what to teach next week, you still prep it in advance, you’re just prepping it quicker.
What happened to me yesterday wasn’t that. I didn’t decide what to teach next week and prep it faster. I decided what my students needed that day and built graduate-level content, the kind that requires real clinical judgment to get right, in the time it took me to walk from my office to the classroom. There was no next week. There was no advance.
That only works because the tool isn’t generic. A general-purpose AI, prompted cold, would have given me something plausible-sounding and probably wrong in the details that matter: the exact threshold, the reasoning chain a clinician uses before ordering a test instead of waiting on the clinical picture. Proctor exists because I spent months feeding it the guidelines, the reasoning, the way I think through these decisions with patients. The slide generator exists because I spent months telling it exactly how I want a deck to look and what a properly formatted clinical image needs. Neither tool wrote my lecture for me. They gave me back my own judgment, formatted and ready, faster than I could have typed it myself.
Faculty saving time is the easy version of this story, the one that shows up in every panel on AI and education. Mine isn’t a time-savings story. I didn’t get my Saturday back. I got rid of the idea that a lecture has to exist before the day you teach it.
I’ve built a lot of tools this year—NursingEdAI, question banks, simulation platforms. Most of them help other faculty do what I did yesterday. But this was the first time one of my own tools changed what I did in my own classroom, in real time, not in a pilot, not next semester. I decided what my students needed and taught it the same day.
I taught something yesterday I hadn’t had time to build. I don’t know how I go back to prepping on a Saturday now that I’ve seen this.