AI has the potential to change industries in ways we’ve never imagined, and healthcare could be one of the most transformed spaces. Between alleviating administrative burden, creating more efficient care pathways, and helping nurses improve patient safety,1 AI can improve the quality and access to care exponentially. Not to mention, agentic systems can help streamline work further, giving clinicians more time to focus on patient care and giving health insurers more opportunities to focus on delivering a seamless and cost-effective experience for their members.2 McKinsey’s latest healthcare leaders survey on gen AI found that 85 percent of respondents have already implemented or are pursuing gen AI use cases in their workflows.3 That number will only rise as this technology advances, making it imperative for health systems to build out their capabilities sooner rather than later to maximize the greatest possible value.
Eric Larsen is president of TowerBrook Advisors and a longstanding expert on the application of AI and other technologies in healthcare. He’s seen how the healthcare industry has both advanced and been reluctant to change over the past 25 years. This moment now, he believes, is the most vital for the industry to jump head-first into change. In this episode of McKinsey on Healthcare, Larsen speaks with McKinsey Partner Jessica Lamb about the ways AI could change healthcare, improve access, and challenge the way we think about care delivery in the United States.
In their conversation, Larsen discusses the tension that incumbents in healthcare and healthcare-focused tech start-ups often have, outlining why incumbents must accelerate their adoption of AI to remain competitive. He describes how the labor-intensive model of healthcare today will be challenged by this technology and the potential of AI to augment already high-performing healthcare teams.
The following transcript has been edited for clarity and length.
Putting incumbents and insurgents in conversation
Jessica Lamb: There are almost two different languages when it comes to the incumbents of the healthcare industry and the new players, such as the tech companies, the AI natives, or the insurgents, as you like to call them. How are you seeing the incumbents and insurgents interacting today? What’s driving the gap between these two entities?
Eric Larsen: The relationship between incumbents and insurgents is important. It’s not incumbents versus insurgents or incumbents or insurgents; it’s incumbents and insurgents. We are going to need both to successfully make this transition. US healthcare has been impenetrable to just about every tech phase shift of the past generation. Internet, mobile, social, cloud, big data and analytics, enterprise SaaS [software-as-a-service], blockchain, and other technologies that have transformed every other vertical in the US economy. I’m trying to bring the incumbents and insurgents into dialogue.
I think about incumbency not as some invariant law of nature but rather as a head start. Depending on your sector within the economy, you have either a little, a lot, or no time to adapt. One thing that’s worth talking about when it comes to AI is the idea of objective, verifiable decision-making. Is there something in healthcare that has a clear right or wrong answer? For example, in revenue cycle management, the claim is either right or wrong—it’s objective. It’s verifiable. And consequently, it will be among the first areas to “agentify.” There’s still a question of how much time the incumbents have to adapt these processes before the insurgents catch up.
Jessica Lamb: As a consultant, I spend the bulk of my time with the incumbents. How do you see competitive tech companies approaching these strategic decisions compared with the healthcare industry broadly?
Eric Larsen: Totally orthogonally. They’re like two separate entities with no understanding of each other. There’s a mutual incomprehension between what’s happening in Silicon Valley and what’s happening in US healthcare. On one hand, US healthcare has been dominated by incumbents. On the other hand, there has been an insurgency of AI and agentic native start-ups. And let’s not forget the hyperscalers. These are the ones that are nation-state-sized in their market capitalizations and in their investable capital. So what is the difference between these insurgents and the incumbents?
They’re like two inverted entities. Institutional healthcare is characterized by its size and heft and permanence—how much revenue they boast and how many tens or hundreds of thousands of employees they have. That’s inverted in Silicon Valley. We’re seeing this phenomenon of tiny teams. By my count, there are some 17 companies in Silicon Valley that have fewer than 50 employees and more than $100 million in annual recurring revenue. The question is, will incumbents figure out how to assimilate and metabolize the most powerful technology in history before the start-ups figure it out? I vacillate on the answer to that.
An industrial revolution like never before
Jessica Lamb: You’re deeply knowledgeable about not only technology but also history, so you bring a unique view to this topic. What should we be learning from previous industrial revolutions?
Eric Larsen: Technology is about advancing in a discontinuous way. History, at least in our retrospective understanding of it, looks linear, but if you decompose it, it’s pretty jagged. We’ve had three industrial revolutions: mechanization, electrification, and computerization. Across those three industrial revolutions, technology uplifted humanity in every dimension, including sanitation, literacy, safety, and democracy. Technology, especially over the last quarter millennium, has marked the upward surge of human flourishing.
Over the first three industrial revolutions, the primary narrative is that the originator of the technology, the inventor, is the winner. So the society that invents the steam engine, or Arkwright’s mill, or the dynamo, or the transistor, or the personal computer wins geopolitically, technologically, in terms of GDP, in terms of total factor productivity, and in terms of advancing their civilizations. But I don’t think that’s the case. To me, the implementation, not the origination, of the technology is what separates the winners from the losers.
In US healthcare, the payers, life sciences companies, or big tech companies that diffuse gen AI into their organizations, even in its current nascent applications for documentation, summarization, coding, and proto-agentic capabilities, will more quickly see the benefits on productivity, lower costs, and access.
Jessica Lamb: This industry has seen hype cycles on the tech side before, so folks are a bit reluctant to jump in with both feet. Why do you believe that this is the most important moment in human history for technology in our industry?
Eric Larsen: I’ll start at an almost cosmic level. Why is this important to civilization? I think about gen AI as a multiplication of intelligence. We’ve based our value on our intelligence. Suddenly, we have created a synthetic intelligence that, in many ways, supersedes human intelligence. Some of the recent model releases have PhD-level understandings of individual disciplines and have encyclopedic knowledge. They know all of humans’ digitized history. With the emergence of reasoning, memory, and tool control, we’re starting to question AI almost like it’s a new species. Is it an invasive species? Is it a benevolent species? There are a lot of different views of this, but my view is that our intelligence, our supremacy, is now being rivaled by a nonbiological intelligence.
In healthcare, gen AI is focused on one thing in this current moment: augmenting productivity and substituting labor for technology. US healthcare has the greatest susceptibility to disruption from this technology for three reasons. First, there are more than 20 million Americans employed in healthcare, and US healthcare is the only industrial vertical to see negative productivity growth over the past generation.
Second, healthcare has an abundance of data. There are 150 zettabytes of data in the world, and one-third of that is in healthcare, increasing 36 percent per year. Now, with natural language processing and vector enablement, a lot of this unstructured data is suddenly usable.
Third, healthcare has a lot of accumulated tech debt, which is a lot of accumulated opportunity, largely because of its imperviousness to past tech phase shifts. So leaders are being confronted with an unprecedented challenge.
Preparing for the emerging tech landscape
Jessica Lamb: What should healthcare executives do in the coming months and years to address this challenge?
Eric Larsen: The first step is to study and understand our current situation and the challenges of employing technology in a highly regulated industry. I’ve described it as “Hippocrates versus Mark Zuckerberg”: “first, do no harm” versus “move fast and break things.” Those cultures collide. That’s one of the reasons healthcare has been able to let technology ricochet through other sectors of the economy. Healthcare patiently assesses tech before it adopts it. Then maybe a stimulant such as electronic health records or transparency or interoperability comes along. But otherwise, we’ve always been able to take our time doing this. I don’t think we’re going to have that time when it comes to AI. So, first, get smart on it.
The second thing is to show vulnerability as a CEO and say to your team, “Hey, I’m figuring this out too.” Show how you’re using AI daily and incorporating it into your workflows to make it psychologically safe for your employees to use. For the past 175 years, we’ve had basically the same hierarchical org chart. Suddenly, we have an augmentation of cognitive labor, which reconceptualizes how we serve our patients and how we organize our incumbent health systems, payers, life sciences companies, and medtechs. It’s important to partner with the insurgents and learn from them so they can help us diffuse this technology safely—and be super thoughtful about data sovereignty and avoiding data exfiltration, especially in a HIPAA environment.
For much of US healthcare, it’s a labor question first. If you as a manager want to hire somebody, you have to convince the CEO that gen AI isn’t going to be able to do this function within a year. There’s a question of going workflow by workflow, occupation by occupation, and understanding if the emerging technology allows us to automate, augment, or eliminate that workflow. Then there’s a fourth imaginative category: Can the technology permit us to do what we couldn’t even conceive of before?
Think about call centers or business process outsourcing, for example. These are multi-billion-dollar industries that I don’t see surviving in a world where conversational AI is ubiquitous. But think about it this way: Clinical start-up Hippocratic has done 3 million calls with patients without a single adverse event. If you had that capability, how many touch points with our patients could we have? In terms of workflows, how can CEOs prime their organizations for the biggest metamorphosis they’ve ever seen in terms of organizing collective human labor?
Jessica Lamb: It’s a huge change, especially for an industry that has not seen much change. What is it going to take to shift from this labor-intensive mode to this capital-intensive mode? What can leaders be doing to prepare their organizations and their teams for what’s coming? How can we make that a productive and positive journey?
Eric Larsen: Since World War II, there has been something of an equilibrium between capital and labor as a share of GDP to the point where it has lulled economists into thinking this was some immutable law of nature. Roughly 70 percent of GDP went to labor, and 30 percent went to capital. Over the past decade or so, we’ve seen a real erosion of that. Now about 60 percent is going to labor, and 40 percent is going to capital. As AI matures and diffuses, you’re going to see an inversion of that number, and the portion will skew asymmetrically toward capital and away from labor. This is going to have massive implications for society and the economy. Every time there’s a big paradigm shift, technology destroys jobs, but it also creates more jobs in counterbalance. With AI, I don’t think that’s going to be the case.
In healthcare, it’s going to play out with more subtlety because we have asymmetric demand and insufficient supply. There are 1.8 million unfilled jobs in healthcare. The first thing to go is that overhang. Societally, I think we’ll see this phenomenon of job hopping turning to job hugging, which we’re already starting to see in tech. Healthcare does not have a static demand. The more affordable you make healthcare, the more it will stimulate demand. Instead of waiting 45 days for a primary care appointment, we could be interacting with our primary care doctors a lot more asynchronously with either them or their avatars. We could see agentic clinical “interns” that expand the reach of our clinicians.
Healthcare is not information technology, but we’ve seen other industries become information technology. There’s only one billion people in the world who have access to a doctor, but there are six billion people in the world who have access to a cell phone. And the cell phone is going to be the form factor for medical super intelligence. We are seeing a proliferation of these medical super intelligences. For example, Epic launched Comet, which was trained on 16.3 billion patient interactions. OpenEvidence now serves 500,000 doctors in the country. There’s Microsoft Diagnostic Orchestrator and GPT-5, which is already superhuman from a differential diagnostic and care treatment and protocol point of view. Turning healthcare into information technology is going to make it deflationary, it’s going to democratize it, and it’s going to make it ubiquitous. I’m very optimistic about that.
AI’s potential to improve access
Jessica Lamb: These are some heavy topics, but you insist you’re an optimist. What are you most hopeful for in the industry in the coming couple of years?
Eric Larsen: I think technology is intrinsically neutral. With the invention of the ship came the invention of the shipwreck. This powerful technology can be used for good or for ill. But I think about the upsides.
A super intelligence that is researching and assimilating millions of peer-reviewed biomedical and scientific studies every year and eventually hypothesizing its own solutions for a new molecule or a new diagnostic instrument can do a lot for medical innovation and civilization. Can it add 30 to 40 years to human longevity? Can it democratize healthcare access across the globe? Can it eliminate the unjustified clinical variation and have “cybernetic” doctors that are omniscient? Can it transform human flourishing and elevate those who are dispossessed or marginalized?
The first step is to have conversations about how to think about all of this. How do we show the incumbents what’s happening from our vantage point and proceed with humility? I feel a consecration, not just a vocation, to our industry and a deep duty to keep learning and share what we’ve learned to partner with the insurgents and break down that division so we can advance this for humanity.
