Thank you for this excellent article. It is not a light read. It takes time and attention to work through the argument. But it is worth every minute. The economic reasoning behind it is powerful, and it points to something that many people in healthcare sense intuitively but rarely articulate clearly.
The central idea is simple but profound. Healthcare has always been organized around scarcity. The scarce resource has been clinician time. Every interaction with the system requires a physician, nurse, therapist, or technician. Since these professionals are expensive and limited in number, the entire structure of healthcare has evolved around rationing access to them.
This scarcity has shaped not only the economics of healthcare but also its culture. We have become accustomed to thinking that more healthcare utilization is a problem. When utilization rises, the reflex reaction is concern about cost.
But the equation you describe challenges that assumption. If artificial intelligence dramatically expands the informational capacity of the system, then the marginal cost of many healthcare interactions falls sharply. When marginal cost falls, consumption increases. That is basic economics.
Other industries have gone through this transition many times. Telecommunications once charged by the minute because network capacity was scarce. Music was sold track by track when distribution was limited. As technology expanded supply, pricing shifted toward access models. Consumption rose dramatically, yet the overall market grew and the structure of the industry changed.
Healthcare has not experienced this dynamic because clinician labor has remained the binding constraint. AI changes that constraint.
What I find particularly interesting is that behavioral health may become one of the first fields where this economic transition becomes visible.
Behavioral health is, at its core, an information service. Most of the work involves conversation, interpretation, pattern recognition, and behavioral guidance. There are no operating rooms involved. The primary instrument of care is the exchange of information between patient and clinician.
Because of this, behavioral health has always been limited by therapist availability. A clinician can see only a fixed number of patients in a week. Sessions are scheduled in discrete blocks. Patients are often seen once every one or two weeks. That rhythm is not determined by clinical science. It is determined by labor scarcity.
From an economic perspective, that structure has always been inefficient. Many behavioral interventions, especially cognitive behavioral approaches, rely on frequent reinforcement. Behavioral change occurs through repeated engagement with new ways of thinking and acting. The ideal therapeutic model often requires more interaction than the system can provide.
Artificial intelligence changes the economics of this interaction.
AI systems can reinforce cognitive frameworks, track symptoms, identify behavioral patterns, prompt exercises, and maintain regular engagement with patients. These interactions can happen daily rather than biweekly. They can occur continuously rather than episodically. And they can do so at a marginal cost that is extremely low compared with clinician time.
Once that becomes possible, the economic unit of care changes.
The traditional behavioral health model is built around the therapy session as the billable unit. A clinician meets a patient for fifty minutes, and the system reimburses that encounter.
But if patients interact with the system every day through digital support, coaching, monitoring, and reinforcement, the concept of the session becomes less central. What matters is not the individual interaction but the ongoing relationship between the patient and the care system.
In economic terms, the industry moves from a metered model to an access model.
Instead of paying for individual encounters, payers begin to pay for continuous support over time. The unit of value becomes the patient over a defined period, often expressed as a per member per month payment.
This is where the argument in your article becomes especially important. When marginal cost approaches zero, the system has room to expand consumption. But the expansion only works economically if pricing models evolve at the same time.
Behavioral health lends itself naturally to these models because its value emerges over time. The benefits of consistent engagement appear in reduced crises, fewer hospitalizations, improved medication adherence, and better functioning in work and family life.
These outcomes accumulate gradually, which makes episodic reimbursement a poor fit. Longitudinal payment models align much better with the nature of the service.
There is also an interesting geopolitical dimension to how this transition might unfold.
In some European healthcare systems with single payer structures, the incentives may actually support faster adoption of these models. When a national health system bears the long term costs of untreated behavioral illness, the economic logic of prevention becomes clearer.
If continuous behavioral support reduces psychiatric hospitalizations, improves workforce participation, and lowers disability claims, the savings accrue directly to the same entity financing the care. The system can justify investing in early intervention because it captures the downstream benefits.
In the United States the situation is more fragmented. Multiple payers, employer turnover, and shifting insurance coverage often dilute those incentives. The organization that pays for preventive care today may not be the one that benefits from the savings five years later.
Yet despite these differences, the direction of travel seems difficult to avoid.
The demand for behavioral health support is enormous. The supply of clinicians remains limited. Artificial intelligence expands the informational capacity of the system while preserving the essential role of human clinicians for complex care.
Once that capacity expands, the economics push the system toward abundance rather than scarcity.
Behavioral care becomes continuous rather than episodic. Payment models evolve toward subscription style access. And the system begins to intervene earlier in the trajectory of illness rather than waiting for crisis.
If the argument in your article proves correct, behavioral health may end up being one of the first places where the concept of abundant healthcare becomes real. Continuous guidance, early detection, and ongoing support would no longer be luxuries available to a small number of patients. They would become the normal operating mode of the system.
For a field that has struggled for decades with limited access and overwhelmed clinicians, that possibility alone makes the economic shift you describe worth paying close attention to.
Thank you for this thoughtful exploration of how AI can transform healthcare economics. The Jevons Paradox framing is compelling, and your examples of Jack versus Jill powerfully illustrate the value of proactive care.
Applying the Tension Transformation Framework reveals an even deeper paradox you've identified but not fully named: the healthcare system's identity-strategy tension is precisely what makes abundant consumption culturally incompatible with current thinking. You write that "consuming more healthcare is bad" is the cultural obstacle—but whose culture are we talking about?
Here's what's critical: the "fear of utilization explosions" you describe is incumbent fear, not patient fear. Health plans, health systems, clinicians, and employers dread utilization explosions because their revenue models are built on scarcity-based pricing. Patients have zero concern about utilization explosions except when scarcity limits their access. Remove the scarcity constraint, and patients would enthusiastically consume infinite healthcare if it improved their health outcomes.
Your pricing models (per task, per workflow, per episode, per patient) are genuinely Creative responses—they redesign the incentive architecture rather than optimize fee-for-service. But here's the structural irony: the very institutions that would need to adopt these models are organizationally invested in the scarcity that made them profitable. The hospital-health plan-PBM industrial complex cannot think their way to zero-marginal-cost infinite healthcare while maintaining their current identity. That's not a critique—it's a diagnostic observation about identity-strategy misalignment.
Utah's AI Sandbox demonstrates what becomes possible when you bypass incumbent identity constraints. Teen mental health support and prescription refill automation—these generate "utilization explosions" with virtually zero incremental cost. They're exactly the innovations incumbents won't pursue because scarcity fuels their revenue models. The sandbox creates space for Architect-identity actors to build solutions the current system is structurally incapable of imagining.
And the demographic reality makes this urgent: clinician shortages are accelerating globally. We're facing a supply-demand imbalance that requires zero-marginal-cost infinite healthcare as the primary model, not a nice-to-have innovation. The current system, operating from Victim identity, will approach this with Maladaptive responses—more regulation to ration access, more consolidation to protect market position, more administrative complexity to maintain scarcity pricing.
The Creative response you're articulating is fundamentally about routing around the incumbent complex. Patients and AI-enabled services need pathways that don't require permission from institutions whose identity depends on scarcity. That's what makes pricing models like "per patient unlimited access" so transformative—they align provider incentives with patient health rather than with utilization management.
The question isn't whether infinite healthcare is economically sustainable. You've demonstrated it is. The question is whether we'll enable Architect-identity actors to build it, or whether we'll allow Victim-identity incumbents to Maladaptively suppress the very abundance that could solve our supply crisis while improving population health. Utah suggests the path forward: create regulatory sandboxes where Creative responses can prove what's possible, then let mobility and federalism propagate the innovations that actually work.
Eventually, everyone cares about healthcare. Perhaps now we can literally afford to care about it earlier in our lives. I'm all for that, especially with aging parents who need more of my time and attention each year.
This was a terrific article. In essence, how do we lower overall costs, improve outcomes and reduce wait times….which all leads to a more productive society.
"What we want to see is the mix of healthcare spending shift away from costly, reactive care and towards proactive, continual service delivery."
I am also expecting that we will see a signifcant shift in healthcare in combination with other innovations, such as in areas like wearables, CGMs, health marker analysis, smart scales etc. Basically, hardware to quantify activity, sleep, weight, glucose, and more already exists and keeps getting cheaper.
Theoretically speaking this would pave the way for creating a prevention focused health insurance structure. One for which you need such hardware and you need to hit regular goals.
Examples include: Regular cancer screenings, weight control, regular workouts, lab cadence for biomarkers...
Policyholders who underperform will receive reduced coverage in those specific areas.
You create an incentive system around longevity optimizations. Great resource for the economic impact of improved longevity: https://silverlinings.bio/
You are left with two key challenges:
1. Regulations. Especially in the US regulation will constrain how hard you can tie coverage to behavior
2. Data Integrity, i.e. what if someone else wears your Whoop for a workout?
The latter can get solved in multiple ways. First of all, the idea is to have comprehensive and regular (for some metrics even real time) data checks. All metrics in combination form the complete picture. If your workout metrics look great but your VO2max and workout tests don't align then you get flagged.
Lastly, a complete endgame solution would be to utilize BCI connections to various smart devices to authenticate as yourself during usage, which can create a cryptographic proof.
Lots of great opportunities here once we are allowed to start thinking about prevention as the main goal of health care / insurance
Also, a huge assumption you’ve made is that medical services actually have an impact on “health”. This is the mythology behind “Healthcare”.
Actually, it’s a things people do on their own, such as smoking less, drinking less, exercising more that are the biggest drivers of healthcare outcomes.
The cost of healthcare services actually have a surprisingly small effect on health.
While more proactive care and increased monitoring has real potential to improve outcomes, more medicine has never automatically meant better health. I don't think the comparison to Telecom or iTunes/Spotify is adequate.
Unlike unlimited streaming, which has zero downside risk, in medicine more diagnostic information can be harmful: false positive and incidentalomas cause patient anxiety and adverse effects from unneeded interventions. This is the argument against full body MRIs for everyone (Prenuvo) and excessive lab testing (Function, etc.). Other examples of when more has not historically been better in medicine: Chemotherapy dose and intensity, PSA screening (for Prostate cancer), Thyroid cancer (see South Korea screening) not resulting in improved outcomes.
The current economic and regulatory structure of healthcare is not built for this kind of change and need addressed. Insurance companies especially are not incentivized to cover prevention services (see "Churn").
It's also a matter of getting this care to those who need it the most, who lack access due to cost, digital and health literacy, and distrust of the system. Additionally, the cost of healthcare services have a surprisingly small effect on health (clinical care counts were only estimated at 10-20% of health outcomes), one reason we are amongst the bottom of developed countries in life expectancy while spending more than any other.
I worry the bottleneck is not the data, but our ability to act on it correctly. For the truly underserved, more access is almost certainly net positive but for the already-served, the marginal benefit of 'unlimited healthcare' is uncertain and the harms are more plausible.
Important piece. As a physician who has already built the model you’re describing, I want to add one layer the essay doesn’t fully address.
You identify two categories: physical interventions that stay expensive, and informational work whose marginal cost AI drives toward zero.
But there’s a third category that matters enormously: interpretive clinical judgment. This does not get commoditized by AI. It gets amplified.
I run a metabolomics-based practice where AI handles clinical synthesis. Holding 500+ metabolite data points, cross-referencing mechanisms across hormonal, mitochondrial, gut, and toxic burden domains, and generating structured clinical documents. The synthesis bandwidth expansion you describe is real. I’ve watched four-hour clinical workups compress to forty-five minutes.
But the value the patient pays for is not the synthesis. It’s the physician who knows which of those 500 data points change the clinical picture, how they interact, and what to do about them in sequence. The AI holds the context. The physician makes the decision. That distinction is the entire business model.
The pricing question you spend most of the piece on, whether per task, per workflow, per episode, per patient, is real for the system-level players. But it may be worth noting that the earliest implementations of the “AI-enhanced physician” model are happening in direct-to-patient, cash-pay practices that stepped outside the reimbursement system entirely.
The patient pays for the interpretation layer directly, and the alignment is clean: they want answers, I deliver answers, no middleman adjudicating whether metabolomics is “medically necessary.”
The future you’re describing is here at the single-physician level, in cash-pay practices, proving the thesis before the system-level infrastructure catches up.
Nice article, enjoyed the read. Along with continuous monitoring and proactive care, we should also stay locked in on developing effective strategies for behavior change. What good is it to know your blood sugar is 200 after eating a pack of double stuffed Oreos, if you intend to do the same thing tomorrow and the day after? What good is it for you to visit your doctor every month and listen to the same discussion on the food pyramid without making a single change in your eating patterns?
Helping a patient change their diet is simultaneously the most simple, most difficult, and most high-yield act a clinician can do. There is a lot of innovation left to in tackling this problem (GLPs, food engineering, AND continuous monitoring).
Is infinite healthcare what we truly want? Or is it infinite health? Most of my patients and most people I know want the latter. Healthcare is good when it's a (the?) pathway to health, not so much when it's not. This sentence in the last paragraph nails it: "The companies that price for abundance rather than scarcity will capture the biggest prize: consumer behavior change, patient engagement, and true health outcomes lift." True health outcomes lift – that's the ticket. AI will multiply productivity many fold in some aspects of healthcare – and yes, those aspects will be used a lot more, but that doesn't mean they will lead to the best outcomes. They will simply lead to the outcomes we're expecting based on the mechanisms we understand. AI is not creating a new vision of what health means or how profoundly we understand a human being and disease. It's simply mining the current atomic paradigms more comprehensively. Our vision, non-mechanical understanding, and subsequent expectations for what is possible form the true ceiling of individual and population health, not compute, which is an intermediate step.
Thanks. Very interesting. I think there is a need for a fundamental intellectual framework for health, medicine and biology. Without that, we're just rearranging the deck chairs on the Titanic. As Stephen Wolfram said in 2024
"Biology is not big on theory. Yes, there’s natural selection. And there’s the digital nature of biomolecules. But mostly biology has ended up just accumulating vast amounts of data (using ever better instrumentation) without any overarching theory. But I suspect that in fact there’s another foundational theory to be found in biology. And if we find it, a lot of the data that’s been collected will suddenly fall into place."
Excellent article! I’d love to see that you address the issue of provider credentialing. Milton Friedman pointed out that medical licensing generally serves to reduce the available supply while having a marginal impact upon the quality of services delivered.
Most of the AI links you provided seemed to provide either “subclinical“ or “support“ services for things like cancer care, etc.
Most medical services are provided through fee-for-service (even today) and these almost always billed under a medical license (MD, DO, RN, PT, PharmD, etc.)
Thanks for the interesting article and hopefully things will work out the way the article described. In my 25 years in healthcare I have never seen insurance companies makes changes because it leads to better treatment, better prevention, or even cost savings. There are many treatments, tests, and processes that could improve things for the patient and the insurer and the insurers have shown no interest. Similarly, patients are reluctant to utilize treatments that can benefit them even if they are free. None of these players act rationally so if and when AI can make improvements in healthcare there will be a lot of barriers. Hopefully, the younger generations will take the lead.
Thank you for this excellent article. It is not a light read. It takes time and attention to work through the argument. But it is worth every minute. The economic reasoning behind it is powerful, and it points to something that many people in healthcare sense intuitively but rarely articulate clearly.
The central idea is simple but profound. Healthcare has always been organized around scarcity. The scarce resource has been clinician time. Every interaction with the system requires a physician, nurse, therapist, or technician. Since these professionals are expensive and limited in number, the entire structure of healthcare has evolved around rationing access to them.
This scarcity has shaped not only the economics of healthcare but also its culture. We have become accustomed to thinking that more healthcare utilization is a problem. When utilization rises, the reflex reaction is concern about cost.
But the equation you describe challenges that assumption. If artificial intelligence dramatically expands the informational capacity of the system, then the marginal cost of many healthcare interactions falls sharply. When marginal cost falls, consumption increases. That is basic economics.
Other industries have gone through this transition many times. Telecommunications once charged by the minute because network capacity was scarce. Music was sold track by track when distribution was limited. As technology expanded supply, pricing shifted toward access models. Consumption rose dramatically, yet the overall market grew and the structure of the industry changed.
Healthcare has not experienced this dynamic because clinician labor has remained the binding constraint. AI changes that constraint.
What I find particularly interesting is that behavioral health may become one of the first fields where this economic transition becomes visible.
Behavioral health is, at its core, an information service. Most of the work involves conversation, interpretation, pattern recognition, and behavioral guidance. There are no operating rooms involved. The primary instrument of care is the exchange of information between patient and clinician.
Because of this, behavioral health has always been limited by therapist availability. A clinician can see only a fixed number of patients in a week. Sessions are scheduled in discrete blocks. Patients are often seen once every one or two weeks. That rhythm is not determined by clinical science. It is determined by labor scarcity.
From an economic perspective, that structure has always been inefficient. Many behavioral interventions, especially cognitive behavioral approaches, rely on frequent reinforcement. Behavioral change occurs through repeated engagement with new ways of thinking and acting. The ideal therapeutic model often requires more interaction than the system can provide.
Artificial intelligence changes the economics of this interaction.
AI systems can reinforce cognitive frameworks, track symptoms, identify behavioral patterns, prompt exercises, and maintain regular engagement with patients. These interactions can happen daily rather than biweekly. They can occur continuously rather than episodically. And they can do so at a marginal cost that is extremely low compared with clinician time.
Once that becomes possible, the economic unit of care changes.
The traditional behavioral health model is built around the therapy session as the billable unit. A clinician meets a patient for fifty minutes, and the system reimburses that encounter.
But if patients interact with the system every day through digital support, coaching, monitoring, and reinforcement, the concept of the session becomes less central. What matters is not the individual interaction but the ongoing relationship between the patient and the care system.
In economic terms, the industry moves from a metered model to an access model.
Instead of paying for individual encounters, payers begin to pay for continuous support over time. The unit of value becomes the patient over a defined period, often expressed as a per member per month payment.
This is where the argument in your article becomes especially important. When marginal cost approaches zero, the system has room to expand consumption. But the expansion only works economically if pricing models evolve at the same time.
Behavioral health lends itself naturally to these models because its value emerges over time. The benefits of consistent engagement appear in reduced crises, fewer hospitalizations, improved medication adherence, and better functioning in work and family life.
These outcomes accumulate gradually, which makes episodic reimbursement a poor fit. Longitudinal payment models align much better with the nature of the service.
There is also an interesting geopolitical dimension to how this transition might unfold.
In some European healthcare systems with single payer structures, the incentives may actually support faster adoption of these models. When a national health system bears the long term costs of untreated behavioral illness, the economic logic of prevention becomes clearer.
If continuous behavioral support reduces psychiatric hospitalizations, improves workforce participation, and lowers disability claims, the savings accrue directly to the same entity financing the care. The system can justify investing in early intervention because it captures the downstream benefits.
In the United States the situation is more fragmented. Multiple payers, employer turnover, and shifting insurance coverage often dilute those incentives. The organization that pays for preventive care today may not be the one that benefits from the savings five years later.
Yet despite these differences, the direction of travel seems difficult to avoid.
The demand for behavioral health support is enormous. The supply of clinicians remains limited. Artificial intelligence expands the informational capacity of the system while preserving the essential role of human clinicians for complex care.
Once that capacity expands, the economics push the system toward abundance rather than scarcity.
Behavioral care becomes continuous rather than episodic. Payment models evolve toward subscription style access. And the system begins to intervene earlier in the trajectory of illness rather than waiting for crisis.
If the argument in your article proves correct, behavioral health may end up being one of the first places where the concept of abundant healthcare becomes real. Continuous guidance, early detection, and ongoing support would no longer be luxuries available to a small number of patients. They would become the normal operating mode of the system.
For a field that has struggled for decades with limited access and overwhelmed clinicians, that possibility alone makes the economic shift you describe worth paying close attention to.
Thank you for this thoughtful exploration of how AI can transform healthcare economics. The Jevons Paradox framing is compelling, and your examples of Jack versus Jill powerfully illustrate the value of proactive care.
Applying the Tension Transformation Framework reveals an even deeper paradox you've identified but not fully named: the healthcare system's identity-strategy tension is precisely what makes abundant consumption culturally incompatible with current thinking. You write that "consuming more healthcare is bad" is the cultural obstacle—but whose culture are we talking about?
Here's what's critical: the "fear of utilization explosions" you describe is incumbent fear, not patient fear. Health plans, health systems, clinicians, and employers dread utilization explosions because their revenue models are built on scarcity-based pricing. Patients have zero concern about utilization explosions except when scarcity limits their access. Remove the scarcity constraint, and patients would enthusiastically consume infinite healthcare if it improved their health outcomes.
Your pricing models (per task, per workflow, per episode, per patient) are genuinely Creative responses—they redesign the incentive architecture rather than optimize fee-for-service. But here's the structural irony: the very institutions that would need to adopt these models are organizationally invested in the scarcity that made them profitable. The hospital-health plan-PBM industrial complex cannot think their way to zero-marginal-cost infinite healthcare while maintaining their current identity. That's not a critique—it's a diagnostic observation about identity-strategy misalignment.
Utah's AI Sandbox demonstrates what becomes possible when you bypass incumbent identity constraints. Teen mental health support and prescription refill automation—these generate "utilization explosions" with virtually zero incremental cost. They're exactly the innovations incumbents won't pursue because scarcity fuels their revenue models. The sandbox creates space for Architect-identity actors to build solutions the current system is structurally incapable of imagining.
And the demographic reality makes this urgent: clinician shortages are accelerating globally. We're facing a supply-demand imbalance that requires zero-marginal-cost infinite healthcare as the primary model, not a nice-to-have innovation. The current system, operating from Victim identity, will approach this with Maladaptive responses—more regulation to ration access, more consolidation to protect market position, more administrative complexity to maintain scarcity pricing.
The Creative response you're articulating is fundamentally about routing around the incumbent complex. Patients and AI-enabled services need pathways that don't require permission from institutions whose identity depends on scarcity. That's what makes pricing models like "per patient unlimited access" so transformative—they align provider incentives with patient health rather than with utilization management.
The question isn't whether infinite healthcare is economically sustainable. You've demonstrated it is. The question is whether we'll enable Architect-identity actors to build it, or whether we'll allow Victim-identity incumbents to Maladaptively suppress the very abundance that could solve our supply crisis while improving population health. Utah suggests the path forward: create regulatory sandboxes where Creative responses can prove what's possible, then let mobility and federalism propagate the innovations that actually work.
Eventually, everyone cares about healthcare. Perhaps now we can literally afford to care about it earlier in our lives. I'm all for that, especially with aging parents who need more of my time and attention each year.
ε(´סּ︵סּ`)з
This was a terrific article. In essence, how do we lower overall costs, improve outcomes and reduce wait times….which all leads to a more productive society.
"What we want to see is the mix of healthcare spending shift away from costly, reactive care and towards proactive, continual service delivery."
I am also expecting that we will see a signifcant shift in healthcare in combination with other innovations, such as in areas like wearables, CGMs, health marker analysis, smart scales etc. Basically, hardware to quantify activity, sleep, weight, glucose, and more already exists and keeps getting cheaper.
Theoretically speaking this would pave the way for creating a prevention focused health insurance structure. One for which you need such hardware and you need to hit regular goals.
Examples include: Regular cancer screenings, weight control, regular workouts, lab cadence for biomarkers...
Policyholders who underperform will receive reduced coverage in those specific areas.
You create an incentive system around longevity optimizations. Great resource for the economic impact of improved longevity: https://silverlinings.bio/
You are left with two key challenges:
1. Regulations. Especially in the US regulation will constrain how hard you can tie coverage to behavior
2. Data Integrity, i.e. what if someone else wears your Whoop for a workout?
The latter can get solved in multiple ways. First of all, the idea is to have comprehensive and regular (for some metrics even real time) data checks. All metrics in combination form the complete picture. If your workout metrics look great but your VO2max and workout tests don't align then you get flagged.
Lastly, a complete endgame solution would be to utilize BCI connections to various smart devices to authenticate as yourself during usage, which can create a cryptographic proof.
Lots of great opportunities here once we are allowed to start thinking about prevention as the main goal of health care / insurance
Also, a huge assumption you’ve made is that medical services actually have an impact on “health”. This is the mythology behind “Healthcare”.
Actually, it’s a things people do on their own, such as smoking less, drinking less, exercising more that are the biggest drivers of healthcare outcomes.
The cost of healthcare services actually have a surprisingly small effect on health.
While more proactive care and increased monitoring has real potential to improve outcomes, more medicine has never automatically meant better health. I don't think the comparison to Telecom or iTunes/Spotify is adequate.
Unlike unlimited streaming, which has zero downside risk, in medicine more diagnostic information can be harmful: false positive and incidentalomas cause patient anxiety and adverse effects from unneeded interventions. This is the argument against full body MRIs for everyone (Prenuvo) and excessive lab testing (Function, etc.). Other examples of when more has not historically been better in medicine: Chemotherapy dose and intensity, PSA screening (for Prostate cancer), Thyroid cancer (see South Korea screening) not resulting in improved outcomes.
The current economic and regulatory structure of healthcare is not built for this kind of change and need addressed. Insurance companies especially are not incentivized to cover prevention services (see "Churn").
It's also a matter of getting this care to those who need it the most, who lack access due to cost, digital and health literacy, and distrust of the system. Additionally, the cost of healthcare services have a surprisingly small effect on health (clinical care counts were only estimated at 10-20% of health outcomes), one reason we are amongst the bottom of developed countries in life expectancy while spending more than any other.
I worry the bottleneck is not the data, but our ability to act on it correctly. For the truly underserved, more access is almost certainly net positive but for the already-served, the marginal benefit of 'unlimited healthcare' is uncertain and the harms are more plausible.
A healthy person has a hundred problems, a sick person only has one. The key to preventative care I believe is in the art of engagement.
Important piece. As a physician who has already built the model you’re describing, I want to add one layer the essay doesn’t fully address.
You identify two categories: physical interventions that stay expensive, and informational work whose marginal cost AI drives toward zero.
But there’s a third category that matters enormously: interpretive clinical judgment. This does not get commoditized by AI. It gets amplified.
I run a metabolomics-based practice where AI handles clinical synthesis. Holding 500+ metabolite data points, cross-referencing mechanisms across hormonal, mitochondrial, gut, and toxic burden domains, and generating structured clinical documents. The synthesis bandwidth expansion you describe is real. I’ve watched four-hour clinical workups compress to forty-five minutes.
But the value the patient pays for is not the synthesis. It’s the physician who knows which of those 500 data points change the clinical picture, how they interact, and what to do about them in sequence. The AI holds the context. The physician makes the decision. That distinction is the entire business model.
The pricing question you spend most of the piece on, whether per task, per workflow, per episode, per patient, is real for the system-level players. But it may be worth noting that the earliest implementations of the “AI-enhanced physician” model are happening in direct-to-patient, cash-pay practices that stepped outside the reimbursement system entirely.
The patient pays for the interpretation layer directly, and the alignment is clean: they want answers, I deliver answers, no middleman adjudicating whether metabolomics is “medically necessary.”
The future you’re describing is here at the single-physician level, in cash-pay practices, proving the thesis before the system-level infrastructure catches up.
Hey — I came across your writing and really liked how you think.
I’m exploring something similar from a different angle — writing about human behavior through a system design lens (like debugging internal patterns).
Just started publishing on Substack. If you ever get a moment to read, I’d genuinely value your perspective.
Also happy to support your work — feels like there’s an interesting overlap here.
Nice article, enjoyed the read. Along with continuous monitoring and proactive care, we should also stay locked in on developing effective strategies for behavior change. What good is it to know your blood sugar is 200 after eating a pack of double stuffed Oreos, if you intend to do the same thing tomorrow and the day after? What good is it for you to visit your doctor every month and listen to the same discussion on the food pyramid without making a single change in your eating patterns?
Helping a patient change their diet is simultaneously the most simple, most difficult, and most high-yield act a clinician can do. There is a lot of innovation left to in tackling this problem (GLPs, food engineering, AND continuous monitoring).
Is infinite healthcare what we truly want? Or is it infinite health? Most of my patients and most people I know want the latter. Healthcare is good when it's a (the?) pathway to health, not so much when it's not. This sentence in the last paragraph nails it: "The companies that price for abundance rather than scarcity will capture the biggest prize: consumer behavior change, patient engagement, and true health outcomes lift." True health outcomes lift – that's the ticket. AI will multiply productivity many fold in some aspects of healthcare – and yes, those aspects will be used a lot more, but that doesn't mean they will lead to the best outcomes. They will simply lead to the outcomes we're expecting based on the mechanisms we understand. AI is not creating a new vision of what health means or how profoundly we understand a human being and disease. It's simply mining the current atomic paradigms more comprehensively. Our vision, non-mechanical understanding, and subsequent expectations for what is possible form the true ceiling of individual and population health, not compute, which is an intermediate step.
Thanks. Very interesting. I think there is a need for a fundamental intellectual framework for health, medicine and biology. Without that, we're just rearranging the deck chairs on the Titanic. As Stephen Wolfram said in 2024
"Biology is not big on theory. Yes, there’s natural selection. And there’s the digital nature of biomolecules. But mostly biology has ended up just accumulating vast amounts of data (using ever better instrumentation) without any overarching theory. But I suspect that in fact there’s another foundational theory to be found in biology. And if we find it, a lot of the data that’s been collected will suddenly fall into place."
Excellent article! I’d love to see that you address the issue of provider credentialing. Milton Friedman pointed out that medical licensing generally serves to reduce the available supply while having a marginal impact upon the quality of services delivered.
Most of the AI links you provided seemed to provide either “subclinical“ or “support“ services for things like cancer care, etc.
Most medical services are provided through fee-for-service (even today) and these almost always billed under a medical license (MD, DO, RN, PT, PharmD, etc.)
Thanks for the interesting article and hopefully things will work out the way the article described. In my 25 years in healthcare I have never seen insurance companies makes changes because it leads to better treatment, better prevention, or even cost savings. There are many treatments, tests, and processes that could improve things for the patient and the insurer and the insurers have shown no interest. Similarly, patients are reluctant to utilize treatments that can benefit them even if they are free. None of these players act rationally so if and when AI can make improvements in healthcare there will be a lot of barriers. Hopefully, the younger generations will take the lead.
Thanks! "Expanding the current market" is a kill for us to pitch the investors so far...