The Opening
Issue two. Last Sunday you let us into your inbox for the first time and didn't leave. Thank you. We don't take that lightly.
A theme is already forming, and we didn't force it. The biggest stories in medicine right now are about a machine learning to do pieces of your job, and doing some of them well. That's not a threat we're here to hype or a panic we're here to sell. It's a shift we want you walking into with your eyes open, because the people deciding how it lands in your hospital are not always the people standing in the room when it matters.
So this week: an AI that out-diagnosed ER doctors in a real test, the human running the biggest bet on ambient AI, four headlines worth ninety seconds, and one thing a TV show got more right than most vendors do.
- Troy, Ray, and Ibrahim
An AI just out-diagnosed ER doctors. Read the fine print.
The phrase to sit with this week is diagnostic reasoning. Not pattern-matching on a scan. Not drafting your note. Actual differential reasoning, taking the same messy, incomplete information you get at triage and working toward the right answer.
A study published in Science this spring put an OpenAI reasoning model (o1-preview) up against that exact task, using real-world data from a Boston emergency department alongside published case challenges. The model didn't just keep up.
Why it matters: On 76 actual ED cases, the model landed the correct diagnosis about 67% of the time at triage, compared with roughly 55% and 50% for the two physician reviewers working from the same limited information. On 143 complex New England Journal of Medicine case vignettes, it included the right diagnosis in its differential 78.3% of the time. This is text in, reasoning out. No labs it didn't have. No hindsight.
The catch: Read the methods before you read the headlines. This was retrospective and text-only. The model was handed clean case write-ups, not a screaming waiting room, a language barrier, and a patient who says "the pain is everywhere." The authors are blunt that it doesn't support replacing physicians, and that the real test is still ahead: prospective, randomized, monitored trials measuring actual patient outcomes and harms. A model that's right on paper is not yet a model that's safe at the bedside.
Bottom line: "AI beats doctors" is the wrong takeaway. The right one is quieter and more useful. A reasoning model can now function as a genuinely strong second opinion, one that never gets tired, never gets anchored, and never stops at the first answer that fits. The question for your shop isn't whether to fear that. It's whether you'll be the one deciding when it gets consulted, and when it gets overruled.
Honest reaction: if an AI flagged a diagnosis you'd missed, what would you feel first?
We're collecting stories. This week, one question.
Every issue we'll ask for one. Here's this week's:
Tell us about the time the technology was right, or the time it was wrong. Maybe an alert caught something you'd have walked past. Maybe a "smart" tool buried the one number that mattered, or charted a detail that never happened. Maybe you're the nurse, the tech, the pharmacist, or the medic who caught the machine before it reached the patient.
We want the real version, not the conference-panel version. Two paragraphs is plenty.
This is open to every corner of medicine: EMS, nursing, pharmacy, techs, registration, environmental services, and administration, all of it. You choose how you're named, whether that's full name, role only, or fully anonymous. We protect patients in every story we run. That is not negotiable and it never will be.
If we run your story, we'll send you a Consult mug as a thank you.
Shiv Rao, MD, Cardiologist & CEO, Abridge
He still takes clinical shifts. That's the detail that matters. Rao is a practicing cardiologist who co-founded Abridge in 2018 on a simple bet: the conversation between a clinician and a patient is the most important moment in medicine, and AI's job is to stay out of the way of it while quietly handling the paperwork it creates.
Where things stand:
Abridge for Nurses went generally available across 250+ health system partners, built for structured flowsheet entry rather than just narrative summaries
Live at Mayo Clinic, Johns Hopkins, Emory, Corewell, Bon Secours Mercy Health and more
Best in KLAS for Ambient AI, two years running (2025 and 2026)
A $300M Series E (mid-2025) pushed the company to a $5.3B valuation
TIME named Rao one of the 100 most influential people in AI
Photo: Modern Healthcare
Why he matters: Most ambient-AI tools were built for physicians. Rao's team is one of the few seriously building for nurses, whose documentation isn't a tidy paragraph but a spreadsheet of discrete fields. If AI is going to give time back to everyone on the floor and not just the docs, this is the harder, more honest version of the problem. And the person steering it still has to live with whatever he ships, one shift a week.
Nurses just got their turn.
Epic's new charting assistant, "Chart with Art," cut end-of-shift note time at Mercy from about 3.5 minutes to roughly 32 seconds per note, an 85% drop, while notes completed fully and on time jumped 225%. Houston Methodist was first to deploy it at the bedside. After two years of ambient AI being a doctor's tool, nursing is finally the headline.
One scan, fourteen red flags.
Aidoc earned FDA clearance for a foundation-model AI that triages 14 critical findings from a single abdominal CT, reported at roughly 97% sensitivity and 98% specificity. It's a marker of where imaging AI is heading: away from one-finding point tools and toward broad models that read a whole study and route the urgent cases first.
House did this every week. Now a model does.
TV’s most famous diagnostician was a fantasy version of exactly what an AI just pulled off in a real ER.
For eight seasons on Fox, Dr. Gregory House solved one impossible case a week by reasoning from a tangle of symptoms to the answer nobody else could see. The character was built on Sherlock Holmes, right down to the name and the apartment number, 221B. And the medical mysteries weren't invented out of thin air. They came from Lisa Sanders, MD, the Yale physician who writes the New York Times Magazine "Diagnosis" column and served as the show's technical advisor.
Here's the thing worth noticing this week. The lone-genius diagnostician, the one who out-reasons the whole room, is exactly the figure an OpenAI model just played in that Boston ER study, landing the right diagnosis in its differential roughly 78% of the time. The difference is that House always had a writers' room and an answer written before the episode started. Real diagnosis runs forward, with missing data and no guarantee the answer even exists.
The show made the brilliant diagnostician the hero. The open question for this decade is whether we still know one when the reasoning comes from a model, and the human's job becomes knowing when it's wrong.
Until next Sunday,
For the people keeping medicine human.
