From Answers to Follow-Through: Where Healthcare AI May Create Its Next Real Value
- 7 hours ago
- 7 min read
Over the past few years, most conversations about AI in healthcare have focused on one question: Can AI understand medicine well enough to help doctors and patients make better decisions?
That question is still important. But I increasingly believe it is no longer the most interesting one.
A recent initiative from the OpenAI Foundation and the Common Health Coalition points toward a different question: Once we know what should be done, can AI help make sure it actually happens?
That distinction may sound subtle. I think it represents a much larger shift in how we should think about healthcare AI.
The initiative, called Breakthroughs to Follow-Through, or B2F, is designed around a persistent problem in healthcare: medical breakthroughs do not automatically translate into health outcomes.
We may already have an effective screening method. We may already know which test should come next. We may already have an effective treatment. And yet patients are still lost between screening, diagnosis, referral, treatment and follow-up.
For me, the significance of this initiative is not really hepatitis C. It is the word follow-through.
Healthcare already knows more than it delivers
One of the strange characteristics of modern healthcare is that our ability to generate information has advanced much faster than our ability to turn information into sustained action.
We can detect more abnormalities. We can produce increasingly sophisticated imaging. We can sequence genomes. We can continuously measure glucose, heart rate, sleep and activity. We can generate increasingly detailed risk assessments. And AI is now making all of this information easier to interpret.
But more information does not necessarily mean better health.
A person may receive a 60-page health screening report containing a lung nodule, elevated glucose, fatty liver, reduced bone density and cardiovascular risk markers. The real problem starts after the report is delivered.
Which issue matters most? What should happen now? What can safely wait? Which specialist should the person see? Does the consultation actually happen? Was the recommended test completed? What happened afterward? Who notices if the person disappears halfway through the process?
Healthcare systems are generally very good at individual episodes of care. They are much less consistent at owning the spaces between those episodes. That is where a large amount of healthcare value is still lost.
The next healthcare AI problem may be execution
The first generation of healthcare AI largely focused on intelligence. Can a model interpret an image? Can it summarize a medical record? Can it retrieve clinical evidence? Can it answer a patient's question? Can it support diagnostic reasoning?
These capabilities matter, and they will continue to improve. But there is a structural limit to an AI system that stops after producing an answer. A recommendation has limited value if nothing follows.
This leads to a different model of healthcare AI: Data → Understanding → Priority → Action → Follow-through → Outcome.
The AI does not need to become the doctor. In many cases, it should not. Instead, AI can become a persistent coordination layer around the patient and the healthcare system.
It can help structure fragmented information, identify what needs attention, help users understand priorities, coordinate the next service, recognize when an expected action has not happened, help patients prepare for the next encounter, maintain continuity across providers and time, and help determine what actually happened after the recommendation.
That is a very different role from an AI chatbot. It is closer to a health orchestration layer.
From AI intelligence to healthcare infrastructure
This distinction also changes how I think about commercial value.
For a long time, healthcare technology companies have competed around access to one important asset: medical knowledge, clinical data, physicians, pharmacies, diagnostics, insurance relationships or patient traffic. But none of these assets alone owns the entire journey.
A hospital may provide excellent treatment but have limited visibility once the patient leaves. A screening provider may identify risk but not manage what comes next. A pharmaceutical company may have an effective therapy but still depend on diagnosis, treatment initiation and adherence. An insurer may finance care but not control the patient experience across hundreds of providers. A large AI model may understand all of these systems conceptually but still have no ability to make the real-world process happen.
This creates a potentially important new layer: healthcare orchestration.
The strategic asset is no longer just data or algorithms. It becomes the combination of Pathway × Provider × Action × Outcome.
In other words: for this type of person, with this type of problem, what pathway should be activated? Which service node is appropriate? Did the action happen? What was the result? Over time, that system can learn not simply from what was recommended, but from what actually worked.
Why this matters beyond the United States
The U.S. healthcare system has its own fragmentation problems, but this challenge is certainly not uniquely American. In China, the problem often appears in a different form.
There is already substantial medical capacity, particularly in major cities. There is a large and sophisticated health screening market. There are extensive hospital networks. There is rapid adoption of digital health tools. And there is enormous interest in medical AI.
But healthcare journeys remain highly fragmented. A person may move between a physical examination center, a tertiary hospital, several specialists, an imaging center, a pharmacy, a rehabilitation provider and a digital health platform without anyone really owning the entire process. Each participant completes its individual task. The patient is left to integrate the system.
This becomes even more obvious in preventive health. We often say that early detection matters. But early detection only matters if it leads to appropriate action.
A positive screening result that never reaches confirmatory diagnosis is not prevention. A risk finding without follow-up is only information. A discharge plan without rehabilitation or follow-up is not continuity of care.
The real opportunity is therefore not simply to create another digital health platform. It is to build infrastructure that helps move people from Finding → Decision → Service → Completion → Outcome.
This is also where longevity becomes practical
This question is particularly relevant to longevity and healthspan.
Longevity is often discussed in terms of biomarkers, advanced diagnostics, biological age, genetics, supplements or emerging interventions. Those areas are interesting. But I believe the more fundamental challenge is much less glamorous: can people consistently make and complete better health decisions over many years?
Most long-term health outcomes are not determined by one extraordinary intervention. They emerge from a sequence of decisions: whether to investigate an abnormal finding, manage metabolic risk, maintain muscle strength, improve sleep, complete rehabilitation, follow up on a cardiovascular risk marker, or return for the next screening.
This is one reason we have been developing NEXA Longevity around the idea of Longevity-as-a-Choice: not longevity as a promise, and not longevity as a number, but longevity as a continuously evolving set of informed health decisions.
Our design principle is that AI should act as a guide rather than an authority, helping people navigate priorities, pathways and choices while professional medical judgment remains where it belongs. That direction increasingly points toward a broader architecture: not only helping people choose, but helping those choices become completed actions.
The loop becomes: Choice → Navigation → Action → Follow-through → Learning → Next Choice.
What this means for AI companies
There is also a broader implication for AI companies. As foundation models become more capable, providing a good answer will increasingly become a baseline capability. The differentiation will move elsewhere.
The question will become: can the AI operate inside a real workflow? Can it interact with real institutions? Can it handle permissions and responsibility boundaries? Can it coordinate humans and software? Can it recover when something fails? Can it recognize when professional intervention is required? Can it track whether the desired outcome occurred?
This suggests that the next generation of valuable AI companies may not simply own the smartest model. They may own the most effective workflow and execution layer around the model.
In healthcare, that means integrating intelligence with care pathways, service networks, human professionals, governance and outcome measurement. The moat may ultimately be less about the model itself and more about workflow + network + trust + outcome data.
From the information economy to the outcome economy
I suspect this pattern will extend far beyond healthcare.
The internet dramatically reduced the cost of finding information. Digital platforms reduced the cost of completing transactions. Generative AI is reducing the cost of producing knowledge, analysis and recommendations. The next step may be reducing the cost of coordinating complex action.
In that world, the unit of value changes. Not: how many answers did the system generate? Not: how many times did the user open the app? But: what actually got done?
A legal AI may not only explain a contract but coordinate the process required to complete the legal task. A financial AI may not only provide analysis but continuously organize the actions required to execute a financial plan within appropriate controls. A healthcare AI may not simply explain a medical report but help ensure that the right next step actually happens.
This points toward what could become an outcome economy. Businesses will increasingly compete on their ability to coordinate fragmented resources around a user's objective. AI becomes not merely an intelligence layer, but a coordination layer between people, institutions and services.
The direction we are watching at NEXA
At NEXA Longevity, this is the part of the B2F initiative that interests me most.
We are not looking at B2F as something to copy. Different health systems require different operating models, different payment mechanisms and different governance structures. What matters is the underlying principle.
Healthcare does not end when the correct recommendation is generated. It ends—or perhaps more accurately, continues—when the right action is completed and its result informs the next decision.
That changes how we think about technology. It changes how we think about health navigation. And it changes what kinds of healthcare companies may ultimately become valuable.
The next major healthcare AI company may not be the company that knows the most medicine. It may be the company that can most reliably connect people, knowledge, decisions, providers, actions and outcomes.
That is a much harder problem. It is also, increasingly, the problem worth solving.
From answers to action. From action to follow-through. From follow-through to better health outcomes.
That is the direction we are watching—and building toward.
Comments