Building the Future of US Healthcare
By Tony Miller Founder and CEO, Harbor Health
After years inside healthcare, working alongside clinicians, employers, and operators, I have come to a clear conclusion: the incentives, not the people, are the real problem.
The United States spends nearly 18% of GDP on healthcare and still underperforms on outcomes that matter. That is not random. It reflects how the system was built.
Fee-for-Service pays for activity rather than outcomes. I have watched capable clinicians operate inside that structure and feel the tension. Reimbursement follows visits, procedures, and billing codes. It does not consistently reward prevention or early intervention to reverse disease progression. Over time, that design produces paperwork, fragmented data, rushed appointments, and steadily rising costs.
When care delivery and coverage are aligned under one structure, the incentives change. Prevention becomes rational. Follow-up becomes built into the model rather than dependent on individual heroics. Costs tend to stabilize because keeping people healthy is no longer financially misaligned.
Kaiser Permanente is the clearest proof that this model works at scale. I have studied Kaiser closely. It is not perfect, but it demonstrates something fundamental. Incentive alignment changes behavior in durable ways.
What has made that model hard to replicate is not the theory. It is the build. Kaiser invested decades in infrastructure, local density, and regulatory positioning to make integration viable. That required patience and capital.
We are operating in a different technological era. Modern data infrastructure lowers coordination costs. Software reduces administrative friction. AI shortens feedback loops that once took months or years to close.
At Harbor Health, we are building an integrated care and coverage model designed for that reality. Insurance and care delivery sit within one aligned structure, so clinicians are not pulled in conflicting directions. We sometimes use the term “payvider”. It sounds like industry shorthand, but the principle is straightforward. Care and coverage function as one system.
Technology is what makes that model scalable today. We think of AI as part of the foundation rather than an add-on. It helps assemble context early, surface relevant signals, and support decisions as care unfolds.
Aligning care delivery and coverage addresses the incentive problem directly. Embedding AI into core workflows gives us leverage. But it only works if clinicians and engineers build it together and if policymakers create space for aligned models to operate responsibly.
If you are a builder, this is one of the most consequential design problems you could choose to work on.
Harbor Health is Building Integrated Care for Today’s Consumer
We believe consumer experience is not a surface feature. It is the operating system.
I do not mean nicer apps. I mean removing the small but compounding frustrations that cause people to disengage. Fewer phone calls. Fewer dead ends. Fewer moments where someone has to repeat their story because context was lost. Less time in stuffy waiting rooms. Scheduling that takes minutes rather than days. Care plans that make sense without requiring specialized knowledge.
We use AI to assemble relevant information before a visit, clarify benefits in plain language, and guide next steps without forcing people to navigate a maze. When the system becomes easier to engage with, people engage. That engagement changes outcomes.
Several design decisions guide our work:
- We reward providers for reducing health risk and total cost of care, not for maximizing visit volume.
- When consumers participate in coordinated primary and specialty care, we remove financial friction. In many cases that means zero out-of-pocket costs.
- Access must be predictable. People should not wait weeks to understand what happens next.
- We intervene earlier in ambulatory settings before issues escalate into emergency visits or hospital stays.
In practice, this model has produced upwards of 20% lower costs than comparable traditional plans. We have honed it in Texas, working closely with employers, clinicians, and local health systems. Today, we serve 250,000 consumers across our plans, including 10,000 in our fully integrated model and 30,000 through risk-based arrangements.
Language Shapes Behavior
We use the word consumer intentionally. The people we serve are not passive recipients of care. They make choices. They compare options. They decide whether to engage.
Our system must respect that agency. If it does not, engagement drops and outcomes follow.
Reducing friction shows up in very practical ways. Most consumers cannot perfectly describe symptoms, and many digital systems require exact terminology. Our AI adapts to how people actually describe concerns, even when spelling or phrasing is imperfect.
Healthcare remains overly dependent on phone calls because digital tools often struggle with complex or ambiguous questions. We are building systems that can handle real-world variability rather than only scripted prompts.
Securing a specialist appointment often involves long waits, unclear coordination, and repeated back-and-forth. Our systems help route consumers to high-performing specialists within our network and coordinate scheduling in a way that reduces that burden.
Benefit documentation is frequently written for compliance rather than clarity. We use retrieval-based models to translate complex plan language into plain English so people can make informed decisions about their care.
The Role of AI in Consumer-Centered Healthcare
Much of today’s AI in healthcare is layered onto legacy workflows. We chose a different path.
We think of AI as part of the infrastructure. It assembles context ahead of visits, suggests next steps grounded in evidence, and learns as outcomes come back so the system improves over t ime. The goal is not automation for its own sake. The goal is better decisions.
We use structured care pathways rooted in clinical evidence. By ingesting clinical, claims, and interaction data, we tailor those pathways to the individual in consultation with their providers.
In many Fee-for-Service environments, a significant portion of each visit is spent reconstructing history instead of making decisions. We generate structured summaries before the appointment begins. That reduces time spent gathering information and allows more time for clinical judgment.
Recently, our system flagged a missed follow-up on a lung nodule that had been noted in imaging months earlier. No appointment had been scheduled. The follow-up caught an earlystage cancer. That kind of quiet intervention is what we are working to make routine.
Most EMR systems are organized around visits and billing cycles. Care does not unfold in discrete episodes. We use the EMR as the system of record while decision support and task orchestration operate in our own layer. That architecture allows us to surface evidence-based actions between visits rather than waiting for the next appointment.
What You’ll Build at Harbor
We have reached a point where the model works operationally across our clinics and insurance products. The next phase is scaling the technical foundation behind it. That requires exceptional engineers and data scientists.
Early on, we chose to build our own data and decision layer rather than let legacy systems define what the model could do. It slowed progress at first, but it reinforced an important lesson. If we wanted a system designed around aligned incentives and continuous care, we needed infrastructure that reflected that philosophy.
Today, our internal ontology maps conditions, care pathways, and progression states, allowing us to build condition-focused models that evolve as care unfolds. That ontology feeds directly into model training and real-time decision support.
If we expect consumers to take ownership of their health, we need builders who take ownership of what they create. Too much healthcare technology has been built around billing workflows and incremental optimization. We are trying to redesign the underlying system.
At Harbor, engineers and clinicians work side by side. You will build probabilistic models that influence real clinical workflows, sometimes catching what would otherwise be missed. You will debate alert thresholds with physicians. You will see the impact of your work reflected in how care teams operate within weeks.
I started this piece by saying the problem is the incentives, not the people. I believe that deeply.
We can build something better. Not by layering new tools onto the old model, but by redesigning incentives and infrastructure together.
This is the work
It is why we started Harbor. It is the standard we hold ourselves to every day.