Why AI Won't Make Your Hospital Special
When technology levels the field, culture becomes the moat
Happy weekend everyone. I have some good stuff this week for you…
The Big Idea — A drill down for the week
Why AI Won’t Make Your Hospital Special
Ideas & Signals — Interesting ideas that I heard that might stretch your mind
Braess’ Paradox from Dr. Graham Walker
News — Things that happened with my take/analysis
ChatGPT5 rolls with a nod to healthcare
OpenAI hints at stewardship of ChatGPT ‘therapists’
Epic could release its long-awaited ambient scribe
Doximity buys Pathway Medical
Dr. Prasad goes to Washington, goes home, then goes back to Washington
Why AI Won't Make Your Hospital Special
As all of us gravitate toward the same tools for patient care, I am thinking alot about what it is that will define care for hospital systems. What will differentiate one hospital from the other? I read Why AI Will Not Provide Sustainable Competitive Advantage in the MIT Sloan Management Review, and it got me thinking how it applies to healthcare.
——
This is happening everywhere: Picture a health system CEO proudly announcing a new AI investment — the most advanced diagnostic model on the market, she says. It cuts CT reading times in half, predicts sepsis hours earlier, and drafts flawless notes for every patient visit. The message to the public and the board of directors? We’re ahead of the curve.
But actually, they're not ahead of anything.
In healthcare, every step in technology from robotic surgery to Epic eventually becomes table stakes. And AI will be no different. The very thing that make so AI powerful will ensure that your competitors are adopting the same tools.
The Homogenization Trap
According to the Sloan Management Review, in business strategy, a sustained competitive advantage must be valuable, unique, and inimitable. AI will always pass the first test but it fails the other two.
In healthcare, homogenization will happen quickly. This is what happens when a breakthrough or tool raises everyone’s game, but erases the gap between them.
Epic offers the same capabilities to every system.
Regulatory guardrails push toward standardized performance.
Shared data networks erode proprietary edge.
Workforce mobility spreads AI fluency.
So, for example, two hospitals both with the same AI-enabled ER triage system will have the same performance. Technology won’t make either stand out.
Residual Heterogeneity
When every hospital has the same CT scanners and diagnostic AI, advantage then comes from how we apply those tools. That’s residual heterogeneity and it refers to the hard-to-replicate qualities (cultural, relational, or creative) that remain in a health system when access to tools and technologies becomes universal.
It might be:
Using freed-up clinician time to accent really unique patient relationships.
Application of the tools for novel care delivery models. Eg, level 2 nursery at home.
Designing unique AI-enabled workflows that reflect your culture
Turning population health insights into proactive outreach programs.
Crushing the things that no algorithm can replicate.
Net net, it's the last point that will define us — if we can recapture a totally unique service experience it will be defining for systems. Quoting a 1996 Harvard Business Review article, this is strategic positioning. It means performing different activities from rivals’ or performing similar activities in different ways.
🔺 AI is the new electricity in healthcare and soon it will be ubiquitous. The winners will be the ones who pair the power of AI with the things that can't be replicated. Aka the people margin.
Ideas and Signals
Braess' Paradox
This is one of the best bits of thinking all week. It's from Dr Graham Walker of Off-Call
In healthcare, we keep widening the highway — then acting shocked when traffic gets worse. Braess’s Paradox explains it in traffic: add a shortcut, everyone piles in, the network clogs, and travel time goes up. If we were ants, maybe we’d coordinate. But we are not ants.
So he's suggesting that when we add capacity and we just create new demand.
The answer isn’t stop building. It’s stop building blindly: Model the whole network before adding a node; Align incentives with flow, not local wins; Measure success in counterintuitive ways (an empty clinic might mean prevention worked).
Read the whole LinkedIn post here. And give Graham a follow.
News
ChatGPT5 rolls with a nod to healthcare. And a patient
OpenAI rolled out ChatGPT5 with attention to healthcare features. Sam Altman said ChatGPT “empowers you to be more in control of your healthcare journey". The official OpenAI launch video (Health discussion begins at 34:59) featured a story of how a patient used ChatGPT5 to understand her biopsy results and regain some sense of agency. The upgrade is just in time — a guy reportedly took some crazy GPT advice on how to control sodium intake and wound up substituting sodium bromide for table salt. He developed bromosis (in fairness, it was under the old 4o). All this with their recent elimination of medical disclaimers and the disarming fact that there's no way to maintain privacy with ChatGPT.
🔺 Sure, we probably need more guardrails. But I think it may be time to build literacy among the user base. We're having a hard time figuring out where this all fits in with our journey. Let Sam do his thing and let's make a concerted effort to discuss and operationalize education for our patients.
OpenAI hints at stewardship of ChatGPT ‘therapists’
But maybe there is some hope: OpenAI is adding new safeguards to ChatGPT to curb unhealthy user dependency by those treating the chatbot as a therapist. The app will prompt users to take breaks during long conversations and avoid giving advice, instead steering toward pros-and-cons lists, etc.
🔺 This shift underscores creeping recognition that generative AI is not just a productivity tool but a real emotional interface, with risks that parallel those seen in mental health. Last week’s letter flagged Sam Altman’s comment that privacy is not protected in these encounters highlighting a huge regulatory gap with these tools. I suspect this wil have to change. | Link
Epic could release its long-awaited ambient-scribe
Politico has reported that Epic will be releasing its own native scribe. The major scribe players have worked desperately to integrate with Epic. Epic has had the chance to see how they work and how they’re used. Now they’re growin’ their own. Christian Pean has a nice summary of the ambient scribe commodity problem.
🔺 The question is what will happen to Suki, Ambience and the rest of the scribe vendors? Recognizing that ambient note-taking is a feature not a product, they have been working desperately to position themselves as platforms that do other things, like coding and billing. But it’s hard to imagine health systems will pay a premium for something Epic will (ultimately) deliver, even if we concede that it won’t be as good. It’s also hard to believe Epic will leave these vendors in the dust. Everyone’s talking pivots, but I don’t know what they can pull out of their hats to keep their subscription seats. And their mega-valuations.
Doximity buys Pathway Medical
Doximity acquired the AI startup Pathway Medical for $63 million. Pathway Medical is an AI-driven clinical decision support (CDS) platform that gives evidence-based answers to clinical questions. It’s like OpenEvidence but leans more toward clinical reasoning rather than citation-based responses.
🔺 Only 23% of primary care doctors are using AI-driven decision support, so there’s a big opportunity to drive stray users back to the Doximity suite. Of course, central to the success of any clinical tool is access and workflow. CDS tools will ultimately be baked-in to Epic.
Dr. Prasad goes to Washington, goes home, then goes back to Washington
Stat is reporting this weekend that Vinay Prasad has been asked back to the FDA after being let go for infuriating the biotech and Duchenne MD community. Not alot of deets on this.
I found this interesting:
Sarepta is the company that had come under Vinay’s regulatory pressure over its gene therapy for Duchenne muscular dystrophy. Seems they hired a Trump-connected lobbyist, Michael Best Strategies. And...
In January, Michael Best Strategies hired Chris LaCivita, who co-chaired Trump’s 2024 reelection campaign along with current White House chief of staff Susie Wiles. CNN reported that Wiles called FDA Commissioner Marty Makary on July 29 to tell him to let Prasad go.
🔺 Conservative/progressive, like Vinay Prasad or not, this is how Washington works. Keep in mind that some of the brightest minds in the business agreed with his assessment of Sarepta’s new drug.
Thank you again for being a subscriber. It means a lot to me. Please pass this along to a friend who might like it.
"Sure, we probably need more guardrails." this is all you can say about the trend for people to load up their privileged health data to parties that do not necessarily only have their interests at heart???