When ChatGPT arrived in late 2022, most of medicine reacted in one of two ways: dismissal or alarm. A smaller group asked a more useful question — how do we actually use this well, and teach others to do the same? That question became Prompt Medicine, one of the earliest structured attempts to bring generative AI into medical education, practice, and research. Three years on, the project has grown from a single book into a course, with a fellowship pathway and a third edition now taking shape. This is the story of that evolution — and an honest account of what the journey has taught me about where AI in medicine is really heading.
Stage OneThe first edition: learning the language
The first edition of Prompt Medicine was, frankly, a book about prompts. That was the right instinct for its moment. In 2023, the models were powerful but brittle; how you phrased a request genuinely changed what you got back. So the book did something practical and unglamorous: it walked through the medical curriculum — anatomy, physiology, pathology, the clinical specialties — and showed, discipline by discipline, how a well-constructed prompt could help a student understand a concept, a clinician structure a differential, or an educator draft a teaching case.
Looking back, the first edition's value was not any individual prompt. It was the demonstration — proof that this technology had a legitimate, structured place in medical work, at a time when much of the profession was still deciding whether to take it seriously at all. It gave readers permission to experiment, and a scaffold to experiment within.
It also revealed the limitation that would shape everything afterward: a book organized around prompts ages at the speed of the tools. The moment a model improves, half the cleverness in your phrasing becomes unnecessary. A guide built on that foundation is always racing its own obsolescence.
Stage TwoThe second edition: from a tool to a landscape
By 2025 the ground had shifted decisively, and the second edition had to be a reimagining rather than an update. Three changes mattered most.
The field stopped being about one model. GPT was no longer the only serious option. Claude, Gemini, Grok, Perplexity, and a growing layer of specialized medical and research tools each brought distinct strengths. The second edition responded with comparative analysis — what each platform does well, where each fails, and which medical tasks suit which system. This was the single most forward-looking move in the book, because it trained readers to choose rather than to depend.
The scope widened beyond the classroom and clinic. New units addressed AI in drug discovery and a structured survey of the AI-resource landscape, alongside deeper treatments of research, medical writing, education, and hospital management. The book grew from five units to seven, and roughly doubled in length.
The tone matured. The second edition leaned harder into a principle I now consider central: the goal is not AI instead of the clinician, but AI that amplifies the clinician's judgment, empathy, and ethical reasoning. The most powerful applications consistently turned out to be the ones where the human stayed firmly in the loop.
If the first edition taught the language of prompting, the second taught something closer to literacy — reading the landscape, weighing tools, and keeping human judgment at the center.
Stage ThreeThe Foundation Module: turning a book into a practice
A textbook can explain. It cannot, by itself, build a skill. That gap is why the Foundation Module of the Prompt Medicine course exists.
The Foundation course concentrates on what every clinician now needs regardless of specialty: core AI literacy and effective prompting. It is the difference between reading about how to verify an AI output and actually doing it, repeatedly, until verification becomes reflex rather than an afterthought. The module is built for the working clinician — the resident drafting a discharge summary, the educator preparing assessments, the consultant who wants to use these tools without being used by them.
Teaching the material rather than merely writing it surfaced a finding I had underestimated: learners do not struggle most with generating AI output. They struggle with judging it — knowing when a confident answer is quietly wrong, when a citation is fabricated, when a plausible differential omits the diagnosis that matters.
Prompting is the easy part. Oversight is the hard part.
That insight is now reshaping the entire project.
Stage FourWhat comes next: advanced modules and a fellowship pathway
The Foundation Module is the first rung of a deliberately designed ladder. Beyond it sit planned advanced modules, each going deeper into a domain where AI is reshaping professional work — advanced clinical reasoning with AI, research and scholarly integrity, medical education, and healthcare leadership and implementation. The intent is not to pile on more prompts, but to build genuine, assessable competence at increasing depth.
The longer arc is to evolve this structure into fellowship-level training — supervised practice, a real-world governance or implementation project, scholarly contribution, and teaching. Medicine already knows how to certify competence through fellowships; the aim is to give clinical AI the same rigor and the same recognized shape, rather than leaving it to ad-hoc self-teaching and tutorials.
Stage FiveThe third edition: beyond prompt engineering
Which brings me to the third edition, now in development — and to a decision that will define it.
After two editions and a course, I have become convinced that organizing the field around prompting, or around individual specialties, is the wrong long-term spine. Every new specialty becomes another near-identical chapter; every named model dates the text within a release cycle. More importantly, it teaches the easy 10% of the skill and neglects the 90% that actually keeps patients safe.
So the third edition is being rebuilt around durable human–AI competencies rather than tools or specialties:
The working subtitle reflects the shift: Human–AI Collaboration in Clinical Practice, Research, Education, and Healthcare Leadership. The market — and the medicine — is moving from prompt engineering toward AI-assisted professional practice. The third edition aims to meet clinicians where the field is going, not where it was when ChatGPT first surprised us all.
Specific tools will live in a clearly dated appendix and a companion resource that can be refreshed as the technology moves. The principles in the body of the book — verification, oversight, ethics, judgment — are built to last, because they describe how to practice, not which button to press.
In closingAn invitation to follow the journey
Prompt Medicine began as a small bet that generative AI deserved a serious, structured place in medicine. That bet has held. But the most interesting part is still ahead: the move from a clever way to query a model toward a genuine discipline of safe, ethical, evidence-based human–AI collaboration.
I will be sharing the development of the third edition openly — the editorial decisions, the competency framework, the course and fellowship design, and the missteps along the way. If you are a student, clinician, educator, researcher, or healthcare leader trying to figure out how to work well alongside these systems, I would value your company on the journey.
The future of medicine will not be written by AI. It will be written by clinicians who know how to work with it — carefully, critically, and well.
Building the third edition — in the open.
Each stage of the journey, from competency framework to fellowship design, shared as it happens. Come build this literacy with me.
Follow the journey at surgeonshamim.com
Loading comments…