AI has moved into medical practice faster than almost anything in modern medicine. As clinicians lean on it more for documentation and clinical decisions, the pitfalls are getting harder to ignore. AI is a powerful bedside tool. It also has myriad ways to screw you. In this episode, emergency physician and clinical AI trainer Sam Ashoo breaks down how to use AI to help you think, how to frame questions to get answers you can trust, what AI scribes are most likely to leave out of the HPI, why clicking "accept" on an AI tool's business associate agreement can put your license at risk instead of protecting it, and how to keep this technology an adjunct rather than a crutch that harms you and your patients.
Guest Bio: Sam Ashoo, MD, is an Emergency Physician and Medical Educator bridging the gap between bedside care and AI. As a clinical informaticist, I explore the future of medicine through a pragmatic, skeptic-first lens to separate clinical signal from hype. He writes at Ashoo Review.
We Discuss:
- What training a medical AI looks like from the physician side
- Which specialties AI is most likely to reshape, and why procedural fields are safest for now
- The biggest misconception clinicians have about AI: it won't tell you what it can't do
- Why omissions, not hallucinations, may now be the bigger risk
- What AI scribes most often leave out of the HPI, including mental health cues and medication changes
- Why AI documentation turns you into an editor, and why most of us aren't good ones
- Treating AI like a smart intern, not the attending, and the rare cases where the reverse makes sense
- UpToDate vs. OpenEvidence: human-curated knowledge vs. machine-synthesized answers
- The trap of asking AI a question you can't judge the answer to
- What "not software as a medical device" means for using OpenEvidence in the ED
- One case, six LLMs, and how changing the question flipped the answer from "send her home" to "cath lab"
- A live OpenEvidence test: is a D-dimer reliable five days after symptom onset?
- Whether to document AI use, and whether your searches are discoverable in court
- Why "I can just ask AI" doesn't replace a fund of knowledge
- How med students, new NPs, and experienced attendings should each use AI
- AI scribe consent: the two steps most clinicians skip, and why all-party consent states matter
- Why you shouldn't upload ECGs or X-rays for AI interpretation
- The HIPAA business associate agreement checkbox that can put your license at risk
Mentioned in this episode:
Doctoring Done Well | Bite-Sized Wins
Every other week, a few minutes of career-elevating insight delivered straight to your inbox.
The Doctoring Done Well Newsletter is never lame, never spammy, and always fresh.5 Free Tools To Make Medical Practice Easier
Scripts for your least favorite conversations.
The quick and dirty guide to calling consults. A 10-minute "Driveway Debrief" to switch off from work. My favorite documentation templates. Step-by-step guide for delivering the news of death.Rob's book now in audio, print, and e-reader versions
Rob's book, Supranormal: A Field Guide for the Impossible Job is now available as an audiobook, narrated by Rob. Find the audiobook, print, and ebook versions at the link below.
Get Out On Time
The Out-On-Time Course gives you the tools to complete your charts on shift, manage overwhelm and interruptions, and create fast, focused, kickass notes, so you can stay out of chart debt and get home on time. Self-paced. 12 hours AMA PRA Category 1 CME.