I thought my AI scribe was the big change. It was the warm-up. AI agents are already in Epic and Abridge, and they don’t stop at the note. As a practicing surgeon, here’s what I want every physician to know before switching one on. I hoped ambient scribes would give me my evenings back. The best data we have says the savings are modest, and after-hours charting barely moved. My real time sink was never the note. It’s the orders, refills, prior auths, and pre-charting that pile up after it. That’s the work agents are built to take over, and it raises a harder question I can’t ignore: who’s accountable when one gets it wrong? Topics -What the largest ambient scribe study to date actually found -The difference between a chatbot and an agent, in plain clinical terms -What Epic (Art, Penny, Emmie) and Abridge are rolling out now -Why reviewing AI drafts is harder than it sounds, and what the anchoring data shows -The single safety question I ask before enabling any agent feature -Whether agents favor physicians or cheaper staffing models -6 agents I'd pay for that no vendor is demoing yet -How agents could make owning a practice realistic again Chapters 0:00 - From AI Scribe to AI Agent 0:58 - Ground Rules: HIPAA and Clinician Review 1:25 - What AI Scribes Actually Deliver 2:36 - Chatbot vs. Agent 3:04 - Epic's Agents and Abridge Orders 3:53 - My Ideal Clinic Day With Agents 4:58 - When AI Mistakes Start Doing Things 5:46 - The Anchoring Problem 6:44 - The One Question to Ask First 7:04 - Utah's No-Signature Refill Pilot 7:27 - Who Are Agents Really Built For? 9:13 - My Agent Wish List 11:08 - Could Agents Bring Back Private Practice? 12:54 - Protect the Time You Get Back 13:49 - Whose Name Is on the Signature? What's the one mundane task you'd hand to an agent tomorrow? Tell me in the comments. The best answers may end up in a future video. Related Videos Utah's AI prescription renewal pilot: https://youtu.be/DylVRYX6YHE My NotebookLM workflow for turning journal articles into a commute podcast: https://youtu.be/eFnKFG5EqWc Sources JAMA ambient AI scribe study (~1,800 clinicians, 5 health systems): https://pmc.ncbi.nlm.nih.gov/articles/PMC13037756/ Lancet Digital Health, GPT-4 drafted patient message replies: https://www.thelancet.com/journals/landig/article/PIIS2589-7500(24)00060-8/fulltext UC San Diego study on AI-drafted in-basket replies: https://www.hcinnovationgroup.com/analytics-ai/generative-ai/news/55405862/uc-san-diego-researchers-study-physician-edits-to-ai-drafted-messages Utah Doctronic prescription renewal pilot: https://commerce.utah.gov/ai/regulatory-relief-4/authorized-pilots/doctronic/ About SpeakMD I'm Dr. Ajay Balaram, a board-certified orthopedic surgeon. I review AI, robotics, and medical devices from inside a busy clinical practice, so you get what holds up in the exam room, not just the demo. New videos for physicians, PAs, NPs, and anyone building in health tech. Subscribe so you don't miss the next one. Disclaimer: For educational purposes only. Not medical or legal advice. Keep patient information inside your EHR or with a vendor under a signed BAA, never in a consumer AI app. Every AI-generated order, refill, or message needs clinician review before signing. #AIinHealthcare #MedTech #PhysicianLife
How AI Agents Will Change the Way Doctors Practice
UpToDate now has an AI layer, and the tier that includes it costs about $600 a year. Before you pay for it, check whether your hospital login already unlocked it. Then see how it stacks up against OpenEvidence and Doximity Ask. I compare all three tools on cost, source material, and drug data, walk through the Nature Medicine head-to-head (including UpToDate's 19% refusal rate), and share how I pick a tool based on the question in front of me. UpToDate Expert AI answers only from UpToDate's own peer-reviewed library: roughly 7,600 authors across 13,000 topics, now with Lexidrug built in for dosing and interactions. I show you how to use it, including the assumptions panel I check before reading any recommendation. CHAPTERS 0:00 - Is UpToDate still worth it in 2026? 1:25 - What UpToDate Expert AI actually is 2:33 - Lexidrug integration 2:58 - How to find and use Expert AI 3:51 - Check the assumptions panel first 5:04 - Case memory and CME tracking 5:30 - Pricing: $579 vs. $699 Pro Plus 6:28 - Why hospitals buy it 7:03 - OpenEvidence 8:10 - Doximity Ask and PeerCheck 9:06 - Head-to-head: cost, sources, drug data 10:33 - The Nature Medicine study 11:17 - The 19% refusal rate 12:40 - My workflow: which tool for which question 14:26 - Ad-funded vs. subscription bias 15:22 - Why the library is the real moat 16:08 - Who wins by 2030 REFERENCED IN THIS VIDEO - Nature Medicine independent head-to-head of clinical AI tools (June 2026) https://www.nature.com/articles/s41591-026-04431-5 - Journal of Medical Ethics study on UpToDate author conflicts of interest (2014) https://jme.bmj.com/content/40/8/578.short www.speakmd.com contact@speakmd.com Never enter identifiable patient information into an AI tool unless it's covered under your institution's HIPAA agreements. Every output needs your clinical judgment and verification before it touches patient care.
UptoDate Expert AI. What Doctors Need to Know.
About a third of American adults used AI for health information last year, and one in six of them used it to decide whether to see a doctor at all. Meanwhile 29% of physicians say no patient has ever disclosed it. Your patients are doing this. They're just not telling you. Discover the implications of artificial intelligence in medical diagnostics and how it affects your practice. In this video, we delve into the scenario where a patient arrives with a pre-existing AI diagnosis, exploring the responsibilities, challenges, and opportunities this presents for healthcare professionals. From verifying the accuracy of AI-generated diagnoses to integrating AI insights into patient care, we discuss the essential steps and considerations for providing the best possible outcomes. Learn how to navigate the evolving landscape of AI in healthcare and enhance your patient care strategies. Whether you're a seasoned medical professional or just starting your career, this video offers valuable insights into the future of medical diagnostics and treatment. Chapters 0:00 - Two Patients, One Word, Two Diagnoses 0:55 - The Nature Medicine Trial Behind It 1:34 - Not a Product Review. A Situation Review. 1:47 - How Many Patients Are Actually Doing This 3:20 - Two Failure Modes 4:24 - Steel Man: Why Patients Turn to AI First 5:07 - Test Results With No One to Explain Them 5:51 - ChatGPT Health Can Now Read the Chart 6:25 - Why Patients Rate AI as More Empathetic 7:22 - Mechanism 1: Incomplete Symptom Input 7:49 - Mechanism 2: Sycophancy and the Patient's Frame 8:56 - Mechanism 3: Liability Guardrails and Test Bias 10:56 - 8 Moves for That Conversation 11:10 - 1. Ask, Because They Won't Tell You 11:42 - 2. Ask What They Typed In 12:06 - 3. Retake the History Out Loud 12:26 - 4. Name the Frame Problem 12:50 - 5. Explain the Testing Bias 13:07 - 6. Give a Better Prompt, Not a Ban 13:28 - 7. Document It 14:30 - 8. If You Do It, Bill For It 14:41 - The Bottom Line www.speakmd.com contact@speakmd.com Sources referenced Nature Medicine, randomized trial on LLM use by patients (2025) https://www.nature.com/articles/s41591-025-04074-y Kaiser Survey https://www.kff.org/public-opinion/kff-tracking-poll-on-health-information-and-trust-use-of-ai-for-health-information-and-advice/ AMA physician survey on AI in practice (~1,700 respondents) https://www.ama-assn.org/practice-management/digital-health/more-80-physicians-use-ai-professionally-ama-survey Meta-analysis of 15 studies comparing AI vs physician response empathy https://pubmed.ncbi.nlm.nih.gov/41115171/
Your Patient Came in with an AI Diagnosis. Now What?
I bought the Plaud NotePin and wore it through a full clinic day and hospital rounds. Here's whether this wearable AI recorder can actually replace your ambient scribe. The NotePin clips onto your white coat and records without you touching your phone. I ran it through my Lisa test, drug names, dosages, lab values, and patient identifiers at full speed, and the AI held up well. But the workflow has real gaps. Nothing transcribes until you sync back to your phone, so you're recording blind through a busy day with no way to tell which recording belongs to which patient. There's no EMR integration, so you're still copying and pasting fragments into the chart. And the business associate agreement only comes with a paid plan, which matters the moment you're touching real patient data. I break down what this device gets right, where it falls short, and who it actually makes sense for. Timestamps 0:00 - Intro: The Problem With Phone-Based Scribes 1:03 - Disclosure & Pricing 1:29 - Unboxing the Plaud NotePin 2:49 - The Pitch: Does It Hold Up? 3:18 - Form Factor & Wearability 4:20 - Alternative Wear Options 5:05 - How the Workflow Actually Works 5:47 - Consent, HIPAA & the BAA Problem 7:27 - The Lisa Test: Real-World Accuracy 9:00 - AI Accuracy vs. Note Structure 9:36 - The Patient-Switching Problem 11:11 - Final Verdict 11:31 - Four Concerns Breakdown 12:38 - Who Should (and Shouldn't) Buy This 13:15 - Wrap Up & More Reviews Plaud AI https://www.plaud.ai/pages/healthcare-solution www.speakmd.com
Plaud NotePin Review. What Doctors Need to Know.
A Lancet study found that experienced endoscopists got worse at finding precancerous polyps after a few months of working with AI. Not while the AI was on. After it was switched off. The ACCEPT trial ran at four endoscopy centers in Poland. Researchers compared unassisted colonoscopies from the three months before AI polyp detection arrived against unassisted colonoscopies from the three months after. Same nineteen endoscopists, each with more than 2,000 procedures behind them. Adenoma detection fell from 28.4 percent to 22.4 percent. Six points absolute, roughly a 20 percent relative decline, on the one metric in that field with a documented link to colorectal cancer mortality. That number sits awkwardly next to a meta-analysis of 44 randomized trials showing AI-assisted colonoscopy raises detection by about 8 percent. The tool works when it's on. This study suggests it may cost you something when it's off. I go through the three mechanisms behind that: cognitive offloading, automation bias, and the quiet loss of unassisted reps. I also spend real time on why nobody should be pulling AI out of their endoscopy suite over one paper. It's observational. Nineteen physicians, four centers, one country, two thin three-month windows. Other GI doctors have already written in to the journal disputing the read. But the mechanism underneath it is one of the most established findings in cognitive psychology, and I tell you where I have caught myself doing the exact same thing with automated documentation. The last stretch is four rules I use to stay sharp, including the one that bothers me most: what happens to a resident who learns with these tools from day one. 0:00 - What Happens When You Take the AI Away 1:29 - Why This Is Not an Anti-AI Video 2:35 - Inside the ACCEPT Trial 4:04 - The Number: 28.4% Down to 22.4% 5:52 - The Contradiction: 44 Trials Say the Opposite 7:16 - Three Reasons the Skill Fades 8:07 - GPS, ChatGPT, and Use It or Lose It 8:43 - Where This Study Falls Short 9:43 - The Pushback in Lancet 10:50 - Where I've Caught Myself Doing It 11:26 - Four Rules for Staying Sharp 12:27 - The Trainee Problem 13:42 - Not the Tool. The Reliance. 15:38 - Coming Next on SpeakMD Study referenced: data from the ACCEPT trial, published in Lancet Gastroenterology & Hepatology, August 2025. https://www.thelancet.com/journals/langas/article/PIIS2468-1253(25)00133-5/abstract Practice Smarter. Not Harder. Questions or collaboration: contact@speakmd.com SpeakMD.com
Is AI Making You A Worse Doctor?
Compounded Ozempic is everywhere, and most doctors have no idea what's actually in the vial. Semaglutide and tirzepatide got compounded and sold for a fraction of the brand-name price during the shortage, and it turned into a multi-billion-dollar market almost overnight. But the FDA shortage is over, Novo Nordisk sued Hims & Hers, and the whole compounded GLP-1 model is heading for a reckoning. I walk through how we got here and what it means for the person sitting across from you in the exam room. I'm Dr. Ajay Balaram, a physician and surgeon. This one hit my radar the hard way. One of my patients had their surgery canceled because they never mentioned they were on Ozempic. They didn't think a mail-order subscription counted as a real prescription. That disclosure gap is exactly what this video is about. What we get into: - What a compounding pharmacy actually is, and how 503A and 503B facilities differ (no jargon, I promise) - How a genuine drug shortage suspended the FDA's "essentially a copy" rule, and why the door never closed when the shortage ended - Why roughly 80% of compounded GLP-1s now have B12 or L-carnitine added, and what that's really about - The Novo Nordisk vs. Hims & Hers lawsuit, and how one ruling could reshape the entire market - The real safety risks: inconsistent potency, sterility in multi-dose vials, foreign-sourced active ingredient, and outright counterfeits - The four questions every patient should ask before they buy CHAPTERS 0:00 What's Actually in Compounded Ozempic? 1:26 Why It Matters for Doctors & Patients 3:01 What Is a Compounding Pharmacy? (503A vs 503B) 5:30 The Shortage Loophole: Why It's So Cheap 8:03 Is It Legal? Novo Nordisk vs Hims & Hers 11:01 Is Compounded Semaglutide Safe? The Risks 12:31 What Good Access Looks Like + What to Ask 13:47 Subscribe & What's Next Physicians: if you take one thing from this, ask the open-ended question at every visit. You can't monitor what you don't know about. And if you're a patient watching, the cheapest option isn't always the safest one, and your primary care doctor should know what you're on. Drop a comment. If you're a doctor, how are you handling patients who show up already on a compounded GLP-1? If you're a patient, what made you choose the route you chose? I read them. SOURCES & FURTHER READING - Omada Health survey of primary care physicians on third-party telehealth GLP-1 prescribing: https://www.omadahealth.com/resource-center/survey-report-primary-care-perspectives-on-glp-1-prescriptions - Novo Nordisk v. Hims & Hers (filed Feb 2026): https://www.novomedlink.com/content/dam/novomedlink/semaglutide/02-09-2026-company-statement.pdf - FDA drug shortage list + compounding guidance (503A / 503B): https://www.fda.gov/drugs/drug-safety-and-availability/drug-shortages CONNECT Subscribe: https://www.youtube.com/@UCJ0cJ1pGnm4z24PIAIl5HVQ More SpeakMD reviews: https://www.youtube.com/@SpeakMD/videos SpeakMD.com This channel is for education and is not medical advice. Talk to your own physician about your specific situation and medications. #Ozempic #Semaglutide #GLP1 #CompoundingPharmacy #Tirzepatide #Wegovy #Telehealth #SpeakMD Practice Smarter. Not Harder.
Online Compounding Pharmacies: What Doctors Need to Know
Everyone is calling Google's new release the AI doctor. Google is going out of its way not to call it that, and the reason why is the whole story. Last week Google DeepMind dropped its AI Co-Clinician, and the headlines wrote themselves: "AI matches doctors." The actual data is more interesting than that. In this breakdown I walk through what Google actually built, the dual-agent architecture running underneath it, and the three studies behind the announcement. I get into the one number every clinician should sit with, which is a single critical error in 98 cases, and what that looks like scaled across a real patient panel. Then I cover the part nobody else is talking about: the gap between the marketing and the medical-legal reality. If you treat patients, that gap is where your liability lives. Chapters 0:00 - The "AI Doctor" Google Won't Call a Doctor 1:24 - Why This Is a Research Announcement, Not a Product 2:27 - The Dual-Agent Architecture (Talker + Planner) 3:00 - What It Can Actually Do 3:48 - The Three Studies: Where Marketing Meets Science 3:57 - Study 1: Evidence Synthesis and the "One Error in 98" Problem 5:08 - Study 2: Pharmacology, the Strongest Claim 5:59 - Study 3: Multimodal Telemedicine Consultations 6:40 - PCPs Win All Seven Domains 7:12 - A Surgeon's Take on the Shoulder Exam Demo 8:03 - No Patient Outcomes, Just Demoware 9:00 - Triadic Care and the Supervision Question 11:21 - The Physician Shortage Argument 12:31 - The Malpractice Reality: Who Holds the Bag? 14:25 - Hallucinations and Sycophancy 15:30 - Three Takeaways 16:42 - Final Thoughts Disclosure This video is for educational and informational purposes only. It is not medical, legal, or investment advice. I have no financial relationship with Google or DeepMind. All product names and trademarks belong to their respective owners. AI Co-Clinician is a research announcement and is not FDA cleared or intended for diagnosis, treatment, or prevention of disease. Google Blog Post - https://deepmind.google/blog/ai-co-clinician/ 📩 contact@speakmd.com #SpeakMD #ClinicalAI #MedTech #GoogleDeepMind #HealthcareAI #PhysicianLife #MedicalLegal #AIinMedicine #DigitalHealth #Telemedicine
Google AI Co-Clinician: What Doctors Need to Know
After a year of using ambient AI scribes across thousands of patient encounters, I've figured out what actually works and what doesn't. I've tested or advised on at least 10 different platforms (DAX, Abridge, Doximity, Iscribe, Nabla, and more), and the gap between what these tools can do and what most physicians are actually getting out of them is huge. In this video, I'm sharing 15 practical tips that will get you the most out of whatever ambient scribe you're using. These come from peer-reviewed research, conversations with colleagues across specialties, and a year of trial and error in my own clinic. Whether you're already using an AI scribe or you've tried one and thought "this isn't for me," there's something in here for you. ⏱️ TIMESTAMPS 0:00 What I've learned in a year of ambient AI scribes 1:46 The 15 tips 2:28 Tip 1: Map it to your phone's action button 3:06 Tip 2: Figure out your consent 3:57 Tip 3: Nail your microphone setup 5:09 Tip 4: Stop talking over your patients 5:50 Tip 5: Chart review out loud in the hallway 7:23 Tip 6: Stop over-narrating in the room 7:54 Tip 7: Verbalize the assessment and plan to the patient 8:40 Tip 8: Dictate the exam as an addendum 9:30 Tip 9: Coach the AI like a junior resident 10:26 Tip 10: Use the structuring keywords 11:23 Tip 11: Customize your template 12:17 Tip 12: Go hybrid with dot phrases 13:31 Tip 13: Verbalize non-verbal findings 13:57 Tip 14: Pause the scribe for sensitive topics 14:34 Tip 15: Never sign without reviewing 15:48 The honest take 16:26 What's next on the channel 📌 MENTIONED IN THIS VIDEO Dr. Christian Pean's article on ambient scribe workflow: https://techysurgeon.substack.com/p/youre-using-your-clinics-ambient 🎥 WATCH NEXT SRT-H, the autonomous surgical robot out of Johns Hopkins Open Evidence breakdown 🔔 SUBSCRIBE for honest, physician-first takes on medical technology. No hype, no sales pitch, just the stuff I'd actually tell my colleagues. 💬 Drop your own ambient scribe tips in the comments. I read them all and I'm always looking to improve my workflow.
What I Learned Using Ambient AI Scribes Daily for a Year
Autonomous robot surgery just crossed a line most people thought was decades away — and as a practicing surgeon, I’m breaking down exactly how an AI-driven robot can now perform a procedure with no human hand on the controls. I cover how systems like the da Vinci platform and the Johns Hopkins SRT-H robot actually work, what “levels of surgical autonomy” really mean, and whether this is safe for patients. If you’ve wondered whether robots will replace surgeons, this is the honest answer from inside the operating room. Timestamps: 00:00 The Headlines 00:56 Welcome to SpeakMD 01:52 How SRT-H Actually Works 04:10 Training Data and Results 05:24 Hype vs. Reality 06:54 The Science Robotics Pushback 09:04 Levels of Surgical Autonomy 10:04 Regulation and Liability 11:17 What This Realistically Means 11:53 Final Thoughts Paper reference: Kim et al., SRT-H: A hierarchical framework for autonomous surgery via language-conditioned imitation learning. Science Robotics, July 2025. https://www.science.org/doi/10.1126/scirobotics.adt5254 If you want physician-first breakdowns of the tech that is actually going to affect your practice, hit subscribe. And if you disagree with anything I said, drop it in the comments. That is how we all get smarter. #SurgicalRobotics #AutonomousSurgery #SpeakMD #MedicalAI #RoboticSurgery #DaVinci #SRTH #MedTech #Intuitivesurgical
Fully Autonomous Robot Surgery with No Human Surgeon Controller?
Subscribe