The Aperture Problem

Shortly after Christmas one year, we noticed that Grandma had lost weight, and that she was struggling with fluid in her lungs. She was feisty and beloved, and she had led a life worth living. Now her path was shifting. The steps remaining seemed fewer, and the family had to figure out what came next.

Even with all of our advanced technology and systems, the work of caring for a patient like Grandma still falls on family. As is commonly the case, Grandpa had passed away years earlier. Thankfully there was a lot of family support, but even so, most of the work fell to one person. My mother stepped up and became Grandma's partner in health.

Over the following months, Mom went with Grandma to her visits to listen, organize, and advocate with the doctors and nurses. She made the appointments. She picked up the medications. She kept watch after Grandma came home from the hospital, and she got her back to the hospital when something wasn't right. She dealt with the insurance company. Grandma had good insurance, but she couldn't drive, she was living with a family member, and there were gaps in understanding and in following the care plan.

Other common social issues came up too. Somewhere along the line, Mom figured out that an insurance salesperson had been calling Grandma and selling her life insurance policies. Five, to be accurate, draining her savings. None of this is the work of the care team. But finances, living conditions, health literacy, nutrition, pets, mobility, and everything else going on complicate the picture. Our care teams simply can't fill the space between appointments, and everything that lives in that space falls to the patient and their family.

After about six months, Grandma passed away, and the family celebrated her life. The hours and minutes of those months were a mix of rewarding and hard. And in many ways the real work began after: the emotions, the memories, a funeral, the assets and possessions.

A lot of people of a certain age have navigated a version of this. Add kids and a career, and it's a lot. It is the unmet burden that healthcare places on families and individuals. Most of that burden is not clinical care. It is coordination. It is rides, paperwork, money, housing, and hard transitions. It craves a different kind of healthcare.

So here is the diagnosis. The foundations of our health system are built on trust and on preventing harm, and rightly so. We anchored everything to scope of practice and licensure as our north star. But that anchor comes with a flip side. Nearly everything in the American health system passes through one aperture: the patient's care team. It is where trust, diagnosis, treatment, payment, documentation, and legal authority all live.

That makes the care team's time the scarcest resource in healthcare. A study out of the University of Chicago put a number on it: a primary care physician would need 26.7 hours per day to deliver the guideline-recommended preventive, chronic, and acute care for a typical patient panel. Shift much of that work to a full care team and it still takes 9.3 hours, more than a workday. The system asks its choke point to do twenty-seven hours of work a day.

We have spent the last two decades building a halo of virtual work around that same aperture. Think of the app portals, the call centers, the population health outreach campaigns for things like cancer screening, the virtual visits and e-visits. These are the right resources. But because of trust and risk, they all stay anchored to the care team, and to the clinician in particular. Patient questions, care coordination, navigation, reassurance, follow-up, prevention: all of it anchors to the care team, and most of it ultimately funnels toward the face-to-face appointment.

The appointment is the one moment the system has the patient's full attention, so we load it with the system's agenda too. We put vitals in there. We put your problem in there. We do preventive health in there. We talk about cancer screenings. We order labs. All of it in twenty minutes, and it really is challenging.

Our EHRs reinforce that structure. The typical electronic health record is designed to maximize the utility of the care team. It is structured as a deep-dive bucket, and it is very transactional, because it drives billing. When the patient is in front of you for twenty minutes, usually ten for the clinician, the team dives fast and deep into that one person's current situation, takes care of the thing the patient came in for, maybe closes a care gap or two, then backs out of the chart, writes a few notes, signs, and dives into the next bucket. Structurally, healthcare is an assembly of shallow, episodic, reactive sick-care moments. The longer arc of a patient's health, their trajectory, their goals, and the non-clinical parts of their life are secondary, or batched into population-level preventive health campaigns. This is not a condemnation. Our EHRs enable incredible work, and I am grateful for our population health tools. Rather, let's name the opportunity and expand our current mindset.

Our problem is an aperture problem. We have a single point of failure, one choke point where we try to do everything for everyone.

And the demand that can't fit through doesn't go away. There is real unmet need that lives outside our four walls, and health systems keep it at arm's length. It lands on someone. In my family, it landed on my mom.

When I started thinking about what an AI-enabled health system might actually look like, I landed on a three-tiered model. I call it the three clocks. It is trajectory-based, longitudinal care: moving from the reactive sick-care that most of our health system is built around today, toward managing a person's health identity over time.

The trouble is that the system runs almost entirely on the fastest of the three.

Clock one is the slow clock. Years to a lifetime. The front end is set by things you did not choose: genetics, epigenetics, inherited conditions, childhood illness, the preventive care you got or didn't. Then your own history starts writing it, where your health has been, what has worked, what hasn't. Your race, ethnicity, income, and neighborhood shape this clock too, not through biology but through exposure and access. Clock one is where goals live. It's the clock you own, and you should be its collaborative author. Our systems do plenty of clock one work — vaccinations, screenings, health maintenance, risk stratification. But it's held as a list of things to complete, not a trajectory to steer. Each item is a gap to close at the next visit.

Clock two runs days to months. This is destabilization. A chronic condition drifting out of range. A hospitalization and the weeks after it. The stretch where something has come loose and hasn't been put right yet. The care team's job here is to carry you through the episode. This is where most suffering and most cost accumulate, and it is where the system is the most manual and disjointed.

Clock three is the fastest. Minutes to days. The acute encounter. You have a cold. Your knee has been bothering you for a while and you finally decide to do something about it, so you try a virtual visit, or primary care, or urgent care. This is real care and it matters, but in the grand scheme of your life it's the grains of rice.

Patients live on clock one, deteriorate on clock two, and are seen on clock three. We have structures that try to manage clocks one and two, but they're anchored inside clock three.

Flip that around and the picture changes. If most of the attention sits in clock one, then clock three still happens — you still get the cold, you still hurt your knee — but now the encounter arrives already informed. The clinician knows the trajectory. They know about the pain management contract, the medications, the advance directives, before anyone starts asking questions.

And near the end of a life, this is what matters most. Instead of care turning reactive and heroic in the final months, the model holds onto the goals that were set when there was time to set them. What are the things you want to spend your time on?

You are the one who is there for the long haul. You should be a bigger owner of the relationship, the goals, and the outcomes.

So how do we get from here to there? How do we shift our health systems toward clock one, and how do we empower patients to be authors?

We are at the beginning of a once-in-a-lifetime technological leap. With AI, the adjacent possible gets very large. There is real potential for calamity. But the more familiar I have become with this technology, the more optimistic I am. That optimism holds only if we adjust our focus, away from efficiency and gain, toward what could be.

Watch the AI councils forming across healthcare and you see both instincts at once. There is an encouraging focus on safety and quality. There is also a gravity that pulls every use case toward making existing systems faster. The most prolific example is ambient documentation, the tools that write the clinician's note. A wonderful tool with great ROI. It is also the perfect illustration of the trap: it makes the aperture more efficient without making it wider. If we limit ourselves to efficiency, we will optimize our way past the unmet need entirely.

To see the other future, a technology has to be able to do four things:

• Hold context.
• Form a lasting relationship with patients and with systems.
• Act across systems.
• Persist over time.

There is a fair objection worth answering here. Every efficiency we have introduced so far has been absorbed. Patient portal messaging did not replace phone calls or visits — it became a third channel stacked on two that never shrank. Free up clinician time and history says it refills.

But the demand I am pointing at is not competing for clinician time today. Understanding your own condition. A simple factual question. What a medication will cost. Who to call next. What happens after discharge. None of that is sitting in a queue waiting for a physician. It goes to a search engine, or to a consumer chatbot, or it lands on a family member at eleven at night. Meeting it somewhere else does not redistribute clinical capacity. It addresses need that is currently unmet, or being paid for in somebody's unpaid hours.

I should be honest that the rest is a hope rather than a mechanism. If we do free clinician time, nothing guarantees it becomes longer visits and better access rather than more throughput. That takes a deliberate decision, and we have not historically made it.

Large language models on their own can't do that. Multimodal agentic AI can, and there are already hints of it. A recent study in Science found that personalized dialogue with an AI durably reduced belief in conspiracy theories, simply by presenting accurate information and answering questions as a trusted partner. The companion robot ElliQ, deployed to older adults through state aging programs, has produced high sustained engagement and reported reductions in loneliness. What both point at is relational technology: something that earns trust over time, which is exactly what a clock one champion would have to do.

I built one of these myself. I have run system-level vaccine programs for six years and watched vaccine misinformation degrade the immune potential of millions of children. So I built a vaccine assistant agent. As a non-coder working with Claude Code, it was an eye-opening experience: partnering with a synthetic intelligence that could do things I couldn't, and coding something trustworthy enough to actually fill a need. The work was building the harness around the model. Boundaries. Vetted content. Escalation paths when someone says something that signals harm. Motivational interviewing built into the flow. A voice that is kind and authoritative at the same time. That harness is where the trust lives.

None of this is a snap of the fingers. The models are further along than the architecture. What does not exist yet is the layer underneath: the orchestration that would let agents hold a trajectory, move across systems, and hand off cleanly to a care team. That is a build, and it is the harder half of the problem.

We need to build this because we are going to need every healthcare worker we can get. AI's job is to support that team, and to take the coordination burden off patients and families — work no paid clinician is doing today. It is being done by my mother, unpaid, or it is not being done at all. That is where the aperture finally widens. AI becomes a member of the team, carrying the load of coordination behind the scenes, and a personal partner in health, helping us co-author our own goals and our own path.

So let's step past the efficiency use cases. Let's start thinking about agents as partners that augment our care teams, so we can redefine our systems to prioritize clock one, build real support for clock two, and let patients be the authors of their own story.

When I establish care with a new clinician, I have an honest conversation with them about how I see myself. I am a lifelong athlete, and my goal is to be active and adventuring into my eighties and nineties. I don't want to be managed against my age cohort. I want to be managed against the ninetieth percentile.

That is what co-authorship looks like from the patient's side. It is also, at the moment, entirely dependent on me saying it out loud to each new clinician and hoping it gets written down.

Trajectory-based care is the model where that goal becomes part of the record instead of part of the conversation. It takes a person's genetics, conditions, injuries, and choices across a life, and works with them toward the best outcome available given the circumstances they are actually in. Much of that is behavioral, and much of it lives outside our walls. In the County Health Rankings model, clinical care accounts for roughly 20 percent of the modifiable factors that determine health. Behavior, social and economic conditions, and physical environment account for the rest. The opportunity is that other 80 percent, and AI can be the bridge that finally brings it into view.

First, the trajectory becomes something the system actually holds. Today every patient has a discrete record inside each EHR they touch. The care team queries it, accesses it by medical record number, makes changes, and leaves. It is manual, state to state, and largely static until someone opens it again. What I am proposing is a trajectory record: not a digital twin and not a risk score, but a lightweight, continuously updated representation of who this person is, where they are headed, and what matters to them. Clinical state, goals, treatment response, social context, and a projected path. The projection is tighter in the near term and more directional the further out it runs. It is a data architecture decision more than an AI problem, and that is what makes it buildable.

There is a second record the system also lacks, and it moves much faster. Call it the coordination record: what is open for this person right now, who owns it, what turn it is on, and what is overdue — across every service line and channel. That is the object whose absence created my mother's second job. Nobody was missing a projection of my grandmother's ten-year trajectory. What was missing was a list of open items with owners and due dates.

Second, divergence replaces thresholds. Alerts today fire when a value crosses a population cutoff. On a trajectory, the same reading means different things depending on where a person has been. Someone with stable diabetes, A1c steady around 7.0, shows a pharmacy fill gap and a three-day glucose shift. No threshold has been crossed. Nothing fires. But the trajectory has diverged, and a low-intensity check-in now costs almost nothing, while the episode it prevents costs a great deal three months later.

Third, the patient co-authors on their own terms. Not everyone can or will participate the same way. Some people contribute simply by living their lives, through wearables, pharmacy fills, the ordinary data exhaust of care. Others confirm and adjust at a handful of moments that matter, a few minutes a year. Some author actively, setting goals and making decisions alongside their team. The system and the care team meet the patient where they are and adapt to the level of engagement and the goals the patient sets.

Fourth, the interface goes where people live. This is the part that could arrive soonest: a relational personal health companion. A collaborative partner that knows you and your family, answers your questions, helps you understand your care, coordinates the pieces, and keeps its focus on long-term wellness. Not another portal to visit. A trusted companion that says, how did your visit go today? I'm looking at your notes. Want me to help you keep the salt down this week? How was Susie's band recital last night? That kind of relationship is built on trust and repetition, and what makes it real is that it aggregates and synchronizes all of your health system records and is truly integrated with your trajectory record.

Underneath all four sits the layer that does not exist yet: a modular orchestration engine that can manage a team of health agents, hold each person's context securely, and interoperate across every system where they receive care. This will take real research, and it will get built piece by piece. But the opportunity is real, and the first piece is a data structure, not a breakthrough.

Everything to this point has been aspirational. It is me proposing one way we could use emerging technology to build a better future for health in America. What's missing is the reason it can't wait.

We already spend an extraordinary amount on healthcare, and a large share of it is waste: repeated testing, and the inefficient seams between all the different partners that make up the system. Researchers have put that waste at roughly a quarter of total spending. But the thing that isn't on the radar is demographics.

Our baby boomers are the largest generation in American history, about 76 million births, roughly 60 percent larger than the generation before them. The oldest of them turn 80 this year. The youngest reach Medicare age in 2029. Over the next twenty years their care needs will climb to levels our system has never seen, and they will stay there.

Here is what matters about the shape of that demand. It is not more visits. Modeling the published utilization rates against the cohorts, what grows fastest is the complex, coordination-heavy end: hospice-type need, post-acute care, multi-condition management. Visit volume barely moves. The fastest-growing category of demand in American healthcare is the category the aperture refuses and hands to families.

At the same time, the workforce that runs the aperture is leaving. About 42 percent of practicing physicians are 55 or older. The AAMC projects a shortage of somewhere between 13,500 and 86,000 physicians by 2036 — a range wide enough to tell you how much is unknown, and one that assumes graduate medical education keeps expanding. If it doesn't, the shortfall runs past the top of the range. No hiring strategy closes a gap that size. No amount of maximizing the aperture compensates for a workforce that will not exist.

And that is the gap measured against how care is used today. The AAMC ran a separate analysis asking a different question: what if underserved populations used care at the same rate as populations with fewer barriers to access? The country would need up to 202,800 more physicians than it has right now, just to meet that demand. More than five times the shortfall projection. The aperture is not only going to be too narrow. It is already too narrow, by a margin no hiring plan touches.

And the third blade, the one almost nobody prices in: the family support network is thinning at the same time. AARP's caregiver support ratio, the number of adults aged 45 to 64 for every person over 80, was 7 to 1 in 2010. It falls to about 4 to 1 by 2030 and below 3 to 1 by 2050. There will be fewer daughters like my mom, and fewer sons like me, to absorb the coordination work the system hands over.

This is not a forecast. It is arithmetic. The people are already alive, and we know how old they are.

Most systems of the kind I've described take two to five years to reach any real scale. Set that against a wave that is already breaking, and the window for starting is not somewhere out in the future. It is now.

So that is the argument. Healthcare hands its coordination work to families because everything has to pass through one narrow opening. Agentic AI is the first technology that can absorb that work, along with the unmet questions and coordination needs of our patients, rather than route around it. And the demographics mean we no longer have the option of waiting to find out.

I don't think I have all of this right. I'm certain some of it is wrong. What I am sure of is that there are a whole lot of well-meaning people, and now agents, out there working on pieces of it in health systems, in startups, in policy forums, and in research groups. I'm writing this to find them.

If that's you, I look forward to hearing your thoughts and ideas, and to building the future together.