The breakpoint of healthcare

When I studied digital media at the IT University of Copenhagen, healthcare was always framed as the holy grail. The unsolvable problem. The Mount Everest of where technology could make a difference. Every other industry had been digitized, optimized, disrupted — but healthcare kept resisting, because healthcare is messy, human, and stubbornly complex in ways that don't fit neatly into software.

And to be fair, when healthcare finally did go "digital," what that mostly meant was taking the paper chart off the shelf and putting it on a screen. Same information, same workflow, now with more clicks. We called that a revolution. The bar was underground. Funny enough, all the little human marks clinicians relied on — the dots, circles, annotations scribbled outside the standard chart — never made it into the digital version. So many implementations actually decreased productivity. But that's another story.

The disruption is no longer coming. It's here.

Matt Shumer wrote a great piece this week that's being shared everywhere. If you haven't read it, the short version: he's an AI startup founder telling his friends and family that the disruption is no longer coming — it's here. That his own job has already been transformed. That yours is next.

He's right. But from where I stand, there are nuances to healthcare that make it fundamentally different from every other industry on his list.

Matt mentions healthcare in one paragraph. He lists it alongside law, finance, and customer service — another industry about to be disrupted. Reading scans. Analyzing lab results. Suggesting diagnoses. Check, check, check.

But healthcare isn't just another line item on the disruption list. It's the Mount Everest that technology has been circling for decades — and AI might be the first thing capable of actually climbing it. At the same time, it's the place where the "AI does your job now" framing falls apart the fastest.

I've spent eight years building electronic health records and four years building digital clinics. I've been in the trenches of healthcare technology long enough to know why it's so hard — and why this moment feels different from anything that came before. And from where I sit, there are two things that are simultaneously true, and you need to hold both of them:

AI could do more good in healthcare than in any other sector on Earth.

And:

AI cannot replace the thing that actually makes people get better.

Let me explain.

The gap AI was born to fill

The healthcare system is broken in ways that AI is perfectly suited to fix. Your doctor has 15 minutes with you. There's no follow-up on whether you changed your diet. Nobody asks why you binge eat at night. Nobody connects your sleep, your stress, your loneliness, and your blood sugar into a single picture — because no human being has the bandwidth to do that for millions of patients.

AI does.

The potential here isn't "AI replaces your doctor." It's that AI fills the enormous gap between the 15 minutes a year you spend with a clinician and the 525,585 minutes you spend on your own. That's where health actually happens. That's where behavior changes or doesn't. And right now, that space is almost entirely empty.

Matt writes about AI handling month-long projects autonomously within three years. In healthcare, the equivalent is an AI that knows your patterns, your triggers, your history — and can guide you through a Tuesday night when you're stressed, alone, and standing in front of the fridge. Not with a generic tip. With something that actually understands your life.

We're building toward that. And the latest models are making it real in ways that would've been unthinkable a year ago.

But here's where it gets more complicated

Matt describes his Monday: tell the AI what to build, walk away, come back to a finished product. That works for code. For healthcare, the picture is different. Any decent model can already tell a person with type 2 diabetes what they should and shouldn't do. That part is easy. But patients are humans. We don't follow instructions out of the blue. Knowing what to do and actually doing it are separated by an ocean of emotion, habit, fear, and context — and that ocean is where healthcare has always struggled.

Healthcare isn't a build problem. It's a trust problem. A vulnerability problem. A "will this person feel seen enough to actually change their life" problem.

Hundreds of millions of people live with cardiometabolic diseases worldwide. Most people on this planet will ultimately die from a lifestyle-related condition. We already know what kills us. We've known for decades. And yet we keep doing it — because we're human. That's not a failure of information. It's a pretty clear sign that information alone was never the only bottleneck.

The best outcomes in chronic disease don't come from better algorithms. They come from someone saying: "I see you. This is hard. Let's figure it out together." That's what the research shows, over and over.

Can AI approximate that? Increasingly, yes. Should AI be the only thing providing it? Absolutely not.

The right model isn't AI replacing clinicians. It's AI handling the 99% of moments where no human is available — and making the 1% of human contact dramatically more informed, more personal, and more impactful.

What this means if you work in healthcare

If you work in healthcare — start using these tools now. Not because AI will replace you. But because the clinician who shows up to a patient visit already knowing that their sleep deteriorated last week, that their mood dipped, that they skipped meals for three days — that clinician is 10x more effective. AI makes that possible.

And it goes further than that. The AI-equipped clinician becomes a turbine — able to grasp thousands of EHR documents, lab results, and patient monitoring data in seconds, cross-referenced against the entire updated lexicon of the latest research and clinical knowledge. Staying current on the best available treatment didn't end at medical school. It didn't end at the occasional CME course at a fancy hotel with a nice dinner. It's now at your fingertips, seconds old, drawn from the latest and greatest evidence available. The clinician who embraces that isn't just more efficient. They're more knowledgeable than any single human could be on their own.

How fast this hits you depends on where you sit. In value-based care models, where outcomes determine revenue, the clinics that adapt slowly will genuinely struggle. The math is unforgiving: if your competitor delivers better results at lower cost because they embraced AI, patients and payers will move.

In public healthcare systems, clinicians can hide behind the red tape for a while longer. But not that long — not once governments realize the productivity and quality gains that are possible. And here's what worries me: when governments do intervene, they tend to do it clumsily. I hope we don't see the classic pattern where a top-down "digital transformation" mandate ends up degrading patient quality while the disruption erodes the trust between clinicians and the system that pays them. That would be the worst of both worlds. The technology is ready to make care better. The question is whether the implementation will be led by people who understand care, or by people who understand spreadsheets.

What this means if you're a patient

There are roughly 8 to 10 million clinical workers missing globally right now. That shortage is why you wait months for appointments, why diagnoses take forever despite requiring nothing more than a series of human touchpoints, why chronic conditions get managed with a pamphlet and a "see you in six months." The system isn't failing you out of malice. It's failing you because there literally aren't enough people.

AI changes that equation. Not by replacing clinicians, but by multiplying what each one can do — and by filling the gaps where no clinician was ever going to show up anyway. That workforce shortage, one of the most intractable problems in global health, could start to dissolve within a few years. Waiting lists that exist purely because of capacity constraints could evaporate if AI is implemented well. This is part of the Mount Everest climb — and for the first time, we have the tools to actually make ground.

What this means if you're building in health tech

Don't just automate the clinical workflow. That's necessary but not sufficient. The real opportunity is the behavior change layer. The space between appointments. The emotional and psychological reality of living with a chronic condition. That's where the massive unmet need is, and that's where AI can do the most good — not by replacing human connection, but by making sure no one has to go through it alone.

The darker side we need to face

But I'd be dishonest if I only talked about the opportunity. There's a darker side to this that we need to face with open eyes.

Every time we integrate these AI models into healthcare, we are handing enormous trust to a small number of foreign, private companies. The models that will soon know your patients' deepest health patterns, their mental state, their behavioral triggers — those models are owned by a handful of labs, mostly in San Francisco. We should be curious about these tools. We should push for them, implement them, build with them. But we should also be clear-eyed about what we're giving away.

The power of these models goes far beyond individual patient care. A technology that can materially improve the health, productivity, and workforce capacity of an entire population is not just a healthcare tool. It's a strategic asset. Countries that implement AI well in their health systems will have healthier, more productive populations. Countries that are locked out — or locked in to a dependency on someone else's infrastructure — won't. This will reshape which nations thrive and which fall behind.

That makes this a matter of national security. Not in the dramatic, military sense — but in the quiet, structural sense that determines which societies are prosperous and stable a generation from now. If we want a world order that stays peaceful, we need to ensure this technology is distributed, not hoarded. We need open access, sovereign infrastructure, and serious conversations about who controls the AI layer that sits between clinicians and patients.

We should be excited. We should also be vigilant.

The GLP-1 pattern, but faster and at scale

There's something else I want to mention, because I think it's coming fast and not enough people are talking about it.

Patients will flood to the path of least resistance. We've already seen this with GLP-1 medications — the moment a quick fix appeared, millions rushed toward it, and an entire industry of telehealth pop-up shops emerged overnight to meet that demand. Many of them operating in a grey zone of clinical practice. Many of them promising top-notch treatment while cutting every corner that makes treatment actually work.

AI-driven clinics will follow the same pattern, but faster and at larger scale. Slick interfaces, instant access, no waiting lists, confident recommendations — everything patients have been starved of. And patients will choose them, because of course they will. When the existing system gives you a six-month wait and a pamphlet, anything that shows up and pays attention feels like a miracle.

But the path of least resistance in healthcare is often not the right course of action. That's the uncomfortable truth about being human — what feels easiest and what actually helps are frequently different things. I predict most of these AI-first clinics will fail their patients. Not because the technology isn't capable, but because they'll optimize for convenience and skip the hard, messy work of real behavior change, real clinical accountability, real care.

It will be a scary live experiment on the health of a generation.

And while governments and payers spend years re-wiring themselves to the new world order of AI in healthcare, it will be the golden age for these pop-up shops. The regulatory gap between what's technically possible and what's properly governed will be wide open, and it will be filled by people moving fast.

But here's where I find hope. This disruption — even the painful parts — will force us to fundamentally rethink what healthcare actually is. Not just who delivers it or how it's billed, but what it means to help someone get better. And from the ashes of the many healthcare systems that will struggle through this transition, something better can emerge. A model that's more human, more continuous, more whole-person — built on what we learned from both the technology and the failures.

That phoenix is worth fighting for.

The breakpoint

And then there's the long-term future — which, at the pace things are moving, might not be that long-term at all.

Elon Musk says it doesn't make sense to study medicine anymore. That's a provocative thing to say. It's also the logical endpoint of the trajectory Matt describes. If AI keeps improving at this rate, there is a future where there are no doctors. Not in the way we mean it today. Robots as surgeons. AI as diagnosticians. Pills and injections prescribed, administered, and monitored without a single human in the loop. The stuff of science fiction — the sterile robotic caregiver — except it's no longer fiction. The pieces are being built right now.

And here's the thing: that future might deliver better clinical outcomes than anything we have today. Fewer errors. Faster diagnoses. Perfect adherence to the latest evidence. No burnout, no ego, no bias. By every measurable metric we use today, it could be superior.

But is that what we want?

There may come a moment — and I think it will come sooner than we expect — where we face what I'd call the breakpoint of healthcare. The point where we have to consciously choose: do we optimize for perfection, or do we choose humanity? Do we want a system that is flawlessly efficient but where no human hand ever touches you when you're scared? Where no one who has lived through illness themselves sits across from you and says, "I understand"?

Maybe we decide the machines are enough. Maybe the outcomes are so much better that we accept the trade-off. But maybe — and this is what I believe — we decide that healthcare was never just about outcomes. That being cared for by another human being is not an inefficiency to be optimized away. It's the point.

That choice is coming. And it will define what kind of society we want to live in, not just what kind of healthcare system we want to build.

Matt ends his piece by saying the future hasn't knocked on your door yet. In healthcare, the future has knocked — most of the industry is just pretending no one's home.

The question isn't whether AI will transform healthcare. It's whether we'll build it in a way that keeps the patient — the whole person — at the center. And whether, when we reach the breakpoint, we'll have the wisdom to choose humanity over perfection.

That's what I'm working on at Embla and hope others will too. And I've never been more convinced it matters.