To those who will remain responsible as intelligence enters healthcare:
IContinuity
A referral can be transmitted successfully and still fail.
A clinician identifies the need. The appropriate order is entered. Supporting evidence is assembled. A health plan applies its criteria. The receiving organization begins its intake process. Messages move, records are updated, and each participant completes the task assigned to it.
The patient may still never receive the intended care.
Nothing necessarily appears broken when each part is examined alone. The clinician acted. The payer responded. The referral was delivered. The receiving institution processed what it received. Every organization may be able to document that its responsibility was performed.
What none of them may be able to explain is why the event as a whole remained incomplete.
The system records completed tasks while the patient lives inside an incomplete event.
We often describe this condition as fragmentation, administrative burden, poor interoperability, or misaligned incentives. Each description is accurate. None reaches the full problem. Healthcare has institutionalized incompleteness. It has become highly capable of processing the fragments of an event without preserving responsibility for the event itself.
A record is not an event. A message is not coordination. A recommendation is not a decision. A decision is not execution. An action is not an outcome. And an outcome that does not change future institutional behavior has not yet become learning.
This distinction explains why healthcare can become more connected without becoming more coherent. Information may cross an organizational boundary without carrying the authority under which it was produced. Evidence may reach a decision-maker without preserving the context that made it relevant. An action may be documented without establishing whether it produced the intended effect. The patient’s outcome may become visible only after the operational and economic systems responsible for the episode have already moved on.
The data continues to travel while the meaning of the event becomes progressively harder to reconstruct.
This is not primarily a failure of individual competence. It is a whole-system failure assembled from locally successful parts. Clinicians, public-health workers, administrators, technologists, researchers, and institutions routinely compensate for discontinuities they did not create and cannot resolve from within the boundaries of their own roles.
The people inside healthcare have often preserved more continuity than the architecture itself.
I have encountered this failure from several sides of the system: complex treatment coordination, research, public-health funding and compliance, surveillance and analytics, health-information governance, and the software through which institutional rules become operational processes. The scale changed. The underlying problem did not.
Healthcare has learned to compute its fragments. It has not yet learned to compute the complete event.
That is the problem from which Vantheon begins.
IIInertia
The last several decades have brought extraordinary progress in healthcare information exchange. Hospitals, laboratories, pharmacies, health plans, public agencies, researchers, and applications can communicate through increasingly mature standards and interfaces. That work is foundational and must continue.
But the movement of information should not be confused with the movement of responsibility.
A message can be delivered without coordination occurring. A form can be completed without establishing that the proper authority existed. A process measure can be satisfied without the patient improving. A billing code can indicate that a service occurred without establishing whether the complete episode succeeded.
Healthcare has therefore developed an economy of proxy truth.
Service volume becomes a proxy for contribution. Process compliance becomes a proxy for quality. Charges become a proxy for cost. Credentials become a proxy for current performance. The successful transmission of information becomes a proxy for a completed handoff.
Each proxy may contain useful information. None is the event itself.
This condition persists because the architecture of healthcare and the economics of healthcare reinforce one another. A clinician may be paid for the encounter, a hospital for the procedure, a laboratory for the test, and a technology company for the interface. Each participant can satisfy the terms by which it is evaluated and compensated while the patient’s underlying problem remains unresolved.
Healthcare does not lack competition. It often competes intensely on the wrong things.
Institutions compete for volume, bargaining power, favorable reimbursement, control of referrals, and the ability to shift cost or risk elsewhere. They do not consistently win by demonstrating better patient outcomes relative to the total resources required across the complete cycle of care.
Value cannot govern a system until value can be established.
A completed service is not necessarily a successful outcome. A guideline followed is not proof that the patient recovered. A charge is not the cost of the resources consumed. The price of one procedure is not the total cost of treating a person’s condition. The relevant outcome may include survival, but it can also include function, pain, complications, time to recovery, treatment burden, the ability to return to work, and whether the improvement endured.
Measurement itself changes the institution. Organizations reorganize around what they are asked to count, reward, and report. A metric can be precise and still be irrelevant to the purpose of care. It can be reproducible and still produce the wrong behavior. A score can improve while the patient’s life remains unchanged.
The misalignment also extends across time. One institution may bear the cost of prevention while another receives the savings years later. A clinical team may invest in better coordination while a different payer benefits from the avoided admission. A public-health agency may create value for people and administrations that do not yet exist.
Even rational institutions become reluctant to invest in outcomes they cannot reliably demonstrate, attribute, or share.
The result is an open-loop system that produces extraordinary volumes of experience but converts remarkably little of it into cumulative institutional intelligence. Exceptions are managed but not learned from. Delays are resolved without improving the next route. A patient may eventually receive care, but the institution remains unable to explain which intervention closed the loop, which friction nearly prevented it, or what should change when the next patient reaches the same point of failure.
Healthcare does not need another isolated intelligence operating inside this architecture. It needs an architecture capable of turning distributed intelligence into coherent institutional action.
IIIAgency
This need becomes urgent as software moves from recording institutional activity to proposing, coordinating, and increasingly performing it.
Artificial intelligence is often described as a technology that will allow people to do more with less effort. That promise is real. But the most consequential technologies have always done more than reduce labor. They have expanded what human beings are capable of creating, understanding, and becoming.
In healthcare, the desirable future is not one in which clinicians disappear from care. It is one in which they are freed from the institutional failures that prevent them from caring as fully as they are capable of doing.
An agent may help follow a patient beyond the limits of a single encounter. A model may interpret bodies of evidence too large for one person to review. A device may reveal deterioration before it becomes a crisis. Public-health institutions may identify patterns across populations with unprecedented speed. Robotic systems may extend skilled physical action across distance and into environments where expertise is scarce.
These systems can enlarge human agency. They can also enlarge the space in which responsibility becomes ambiguous.
Clinical software can already de-identify information locally, structure a case, formulate a plan, invoke specialized tools, calculate risk, and generate a proposed disposition. This is meaningful progress. It suggests that useful clinical computation can increasingly be composed across multiple models and tools while reducing the unnecessary movement of identifiable information.
But a completed sequence of model calls is not a completed clinical event.
The software may know which text it received, which tools it invoked, and which disposition it produced. The institution must still establish whether the source evidence was complete, whether the transformed information remained clinically sufficient, whether the proper rule was applied, whether the agent possessed authority, whether a legitimate decision-maker accepted the recommendation, whether the intended action occurred, and whether the patient achieved the expected outcome.
A disposition is produced inside a computational environment. An outcome occurs in the life of a person.
Frontier-model evaluations are necessary, but they cannot answer these institutional questions. A model may be sufficiently safe to enter the market and still be inappropriate for one patient. It may perform well on national benchmarks and remain unauthorized for a particular action. It may be generally reliable and still lack the local evidence, contractual scope, clinical delegation, or public mandate required in one institution at one moment.
A model may be qualified to exist without being authorized to act.
General intelligence is designed to generalize. Healthcare legitimacy remains local. It resides in the relationship among a particular rule, the evidence available at the time, an identified actor, a legitimate authority, an action, and the outcome that followed.
The urgency increases when intelligence enters the physical world. Last week (July 2026), researchers reported that supervised humanoid robots had performed laparoscopic tasks in a live preclinical setting. Human surgeons remained in control; the work was not autonomous surgery and was not conducted on human patients. Its institutional importance nevertheless extends beyond the specific experiment. Physical execution had been distributed across professional judgment, remote control, software, communications infrastructure, robotic embodiment, instruments, and the institution in which the event occurred.
The milestone was not that a robot had become a surgeon. It was that physical execution had become separable from professional authority.
Healthcare institutions know how to credential a clinician and register a medical device. They are less prepared to preserve a continuous account of authority, evidence, responsibility, and consequence when cognition, control, embodiment, and outcome are divided among human and machine participants.
Robots may hold the instrument. The institution must still hold the truth.
IVCoherence
Vantheon exists to make the complete healthcare event computable without transferring institutional authority to the machines that participate in it.
We are building a learning, verification, and settlement infrastructure for regulated health systems. Its fundamental unit of computation is not the isolated message, claim, authorization, model output, or transaction. It is the complete institutional event.
An institutional event is the governed relationship among the applicable rule, the evidence available, the authority to act, the decision made, the execution that occurred, the outcome that followed, and the operational or economic state that should come next. It must also preserve provenance, uncertainty, exceptions, and the possibility of review.
Consider the referral with which this letter began. In most systems, the referral appears complete when the message is transmitted. For the patient, it is complete only when the receiving organization accepts responsibility, the required evidence is available, the service occurs, the result is understood, unresolved obligations are addressed, and the originating institution knows what happened.
Vantheon is designed around that difference.
We describe its institutional capacities as Learn → Verify → Settle. In the life of an individual event, verification establishes what can be supported. Settlement closes what follows. The resulting learning changes how the next event is governed.
Verification connects the rule to the evidence, the evidence to authority, authority to execution, and execution to the observable outcome. It asks not only whether a task occurred, but whether the participant was authorized, whether the evidence was sufficient, whether the action remained within scope, whether the governing rule was current, and whether the claimed result can be supported.
Verification cannot assume that a structured record is true merely because it is digitally signed or transmitted through a conformant interface. Evidence may be incomplete, contradictory, outdated, or produced for a different purpose. A model may be calibrated for another population. An instrument may be unreliable. The verification process itself may contain error.
For that reason, Vantheon must preserve more than conclusions. It must preserve context, provenance, uncertainty, contradiction, version history, and the conditions under which a conclusion remains valid.
Learning begins when an institution can use what happened to change what it does next. A completed event should reveal which pathways worked, which exceptions recurred, where authority failed, what evidence proved decisive, which resources were consumed, and which conditions shaped the outcome.
The institution should not merely retain a record of the event. It should become better because the event occurred.
That learning must remain governed by the institution that produced it. Models, vendors, applications, and technical architectures will change. The accumulated intelligence created through an institution’s own execution should survive those changes.
A healthcare institution should not have to rent its own memory.
Settlement closes the institutional state created by the outcome. An event is not complete because an action was documented, a referral was sent, or a procedure ended. It becomes complete when the result is understood, responsibility remains visible, unresolved obligations are identified, exceptions are addressed or escalated, and the appropriate operational or economic consequence can be calculated, reconciled, or recommended.
Settlement may eventually involve payment. It begins before payment, when the event can no longer disappear between systems.
During our foundational work, we have defined the complete-event architecture, the ontology that connects its institutional elements, the Codex through which rules and authority can remain explicit and versioned, and the trace required to connect execution with outcome, learning, and reconciliation.
This is not a claim that healthcare has already been transformed. It is the work required before such a claim can become credible.
We are now entering our first operating phase. We will begin with bounded coordination events in which evidence, authority, execution, institutional handoffs, and remaining obligations can be made visible as one traceable whole. We are not beginning by attempting to automate all clinical judgment, replace the electronic health record, create an unquestionable universal trust score, move funds without authorization, or resolve medical and contractual disputes through software alone.
The initial objective is narrower and more demanding: preserve one consequential event across its institutional boundaries; establish which evidence and authority governed it; identify exceptions; determine what occurred; produce an auditable record; and make clear what should follow.
The distance between an architecture and an operating institution is evidence. We intend to cross that distance one complete event at a time.
VLegitimacy
The construction of this infrastructure creates questions that cannot be hidden behind the language of responsible innovation.
A system capable of making accountability more precise can also make injustice more efficient. A performance model can punish clinicians who serve more complex populations. A technically verified rule can still encode an unjust policy. A settlement mechanism can automate an incentive that should never have existed. Infrastructure designed to make execution visible can become an instrument of surveillance.
Mathematical precision does not create institutional legitimacy.
The fundamental question is not only whether Vantheon can compute accurately. It is who has the authority to define the rule, determine what evidence is sufficient, establish what constitutes a meaningful outcome, and decide which consequence is justified.
The answer cannot be the model, the robot, the vendor, or Vantheon acting alone.
Legitimate clinical, public, contractual, administrative, and legal authority must remain where institutions and democratic processes have placed it. Vantheon’s role is not to manufacture authority. It is to make authority explicit, versioned, inspectable, and connected to the event for which it is responsible.
Consequential institutional logic must also remain contestable. A person or organization affected by a determination should be able to understand which rule was applied, which evidence was considered, which system version participated, where uncertainty remained, and how the result can be reviewed.
Technology does not merely automate an institution. It redistributes power within it.
Whoever controls the interface through which evidence is presented, the models through which it is interpreted, the pathways through which action is permitted, and the infrastructure through which consequences are reconciled can shape institutional behavior without formally appearing to govern.
A platform can begin as a useful service and gradually become a private jurisdiction over the organizations that depend on it. An institution may formally retain authority while losing the practical ability to understand its own operations, replace its providers, recover its accumulated intelligence, or act outside pathways established by the platform.
Vantheon must not become such a system.
The institutions we serve must retain meaningful control over their rules, evidence, delegations, models, records, and learning. Models must be replaceable. Decision histories must remain portable. Consequential determinations must remain reviewable.
An institution should not have to rent its memory. It should not have to rent its capacity to act.
Nor should we assume that decentralization alone resolves the problem. A distributed network can still contain concentrated economic or governance power. A transparent ledger can faithfully record the execution of an illegitimate rule. An automated process can reproduce an unjust incentive with greater consistency than any human administrator.
Decentralization is an engineering property. Legitimacy is an institutional achievement.
These principles must constrain our business, not merely our language. We should create value because healthcare events become easier to complete, understand, and learn from—not because the cost of leaving our infrastructure becomes unbearable. We should not profit from ambiguity, inaccessible intelligence, duplicated work, or permanent institutional dependence.
We should not profit from keeping the loop open.
Those who provide capital to Vantheon should understand the company they are being asked to own. We will prefer durable institutional infrastructure to demonstrations designed primarily to attract attention. We will prefer verifiable adoption to the appearance of rapid scale. We will narrow claims when evidence is incomplete and refuse uses when authority and accountability are unclear.
Those who build Vantheon should understand the standard by which their work will be judged. Completing a document is not the same as resolving a question. Connecting an interface is not the same as closing a workflow. Producing a model output is not the same as establishing a decision. Recording an action is not the same as understanding its result.
We will not hand ambiguity from one function to another and call the transfer progress.
Trust cannot be a marketing asset for a company building infrastructure for trust. It must be an operating obligation.
VIFulfillment
At a civilizational scale, health is more than the absence of disease. It is the capacity of individuals and populations to pursue lives of meaning, autonomy, participation, and fulfillment.
Medicine protects and restores that capacity through science and care. Institutions determine whether that science and care can reach people coherently, fairly, and in time.
The healthcare system of the future will contain forms of intelligence and agency that earlier generations could not have imagined. Clinicians will work with agents capable of coordinating across vast bodies of knowledge. Devices will observe continuously. Robots will extend skilled physical action. Public institutions will understand patterns across populations with unprecedented speed and precision.
The existence of these capabilities does not guarantee that they will improve human health. Their value will depend on whether institutions can coordinate them without dissolving individual sovereignty, professional responsibility, public accountability, or the meaning of care itself.
The future should be materially better than the system we have accepted.
A patient should not have to serve as the courier of evidence among organizations. A clinician should not have to reconstruct the same institutional history during every encounter. A public-health team should be able to change a rule and understand how that change moves through execution. An institution should be able to observe an outcome, understand why it occurred, and improve before another person reaches the same point of failure.
Greater intelligence should return time and capacity to care, judgment, discovery, and human presence. It should not create another layer of distance between people and the institutions responsible for them.
Reaching that future will require three enduring capacities.
The first is sovereignty: meaningful authority over health decisions, participation, data, and consequence.
The second is coordination: the ability of clinicians, institutions, agents, devices, and robots to participate in one coherent event without allowing responsibility to disappear between them.
The third is learning: the ability to convert verified outcomes, exceptions, failures, and discoveries into better future judgment while preserving ownership of that intelligence within the institutions and communities that produced it.
Together, these capacities define programmable healthcare.
Programmable healthcare does not mean that software determines what human life is worth. It means that institutional rules, delegations, evidence requirements, actions, outcomes, and obligations become explicit enough to execute, inspect, contest, and improve.
It must also remain capable of change. Medical evidence will evolve. Public mandates will change. Contracts will change. Models, devices, and institutions will change. Rules must therefore be adaptable without erasing the versions that governed prior events. Learning must alter the future without silently rewriting the past.
A system that cannot change will become obsolete. A system that changes without memory will become ungovernable.
Vantheon is being built as infrastructure for that future: a system through which health institutions can learn from verified experience, establish the truth of execution across human and machine participants, and reconcile the consequences that follow.
We do not believe that civilization’s most advanced healthcare system will be the one containing the greatest number of intelligent machines. It will be the one most capable of directing intelligence—human, artificial, and embodied—toward the expansion of human possibility.
Medicine has spent centuries enlarging what can be done for the human body. The work before us is to build institutions capable of learning from that power, governing it responsibly, and extending its benefits to more people.
The future of healthcare will be programmable. The decisive question is who will have the power to define its rules, shape its behavior, possess its intelligence, and claim the value it creates.
It can be programmed around transactions, dependency, institutional convenience, and machine capability. Or it can be programmed around human fulfillment, legitimate authority, individual sovereignty, verified outcomes, and shared learning.
We have chosen our answer.
That is why we are building Vantheon.