10,000 Patients Is Not the Answer

Today I saw a post from the CEO of a healthcare software company that got me a little riled up.

He was reporting on a debate that broke out at an industry conference. The question on the table, in his words: "whether it's great for healthcare if 1 family doc uses AI agents to manage a 10,000-patient panel OR is that the DEATH of primary care if the PCP doesn't personally know every patient?"

The post was illustrated with a cartoon. On the left, 1930: a doctor in a hat and overcoat climbing the front steps of a house, black bag in hand, a woman holding the door open for him. On the right, 2030: a grid of identical houses receding to the horizon, and in front of them a doctor at a desk with a laptop, ringed by floating video tiles of patients' faces.

The caption reads: "In 1930, the doctor came to your home. In 1980, you went to the doctor. In 2030, AI may let one doctor come to 10,000 homes again."

Here's the detail that kept me chewing on this all day: the person who posted it is a physician. Dr. Joshua Liu holds an MD from the University of Toronto and is co-founder and CEO of SeamlessMD. He does not appear to have a clinical practice.

I don't raise that to question his credentials. I raise it because of what his company was built on.

The origin story, as SeamlessMD tells it, is that during his medical training he worked on research into hospital readmissions — patients who go home and come back, because something got missed in the space between the people caring for them. The company he went on to build exists to close that gap. Its entire premise is that continuity of care is not a nicety: that when the thread between a patient and the people looking after them gets thin, people get hurt, and it shows up in the data.

I agree with that premise completely. What I can't work out is how you hold it in one hand and a 10,000-patient panel in the other.

Because a panel of 10,000 is that thinning. Engineered deliberately, and at scale. It is the thinnest possible thread between a patient and a physician — a name on a chart, an agent in the middle — and we are being invited to consider whether it might be good.

It is also very difficult to hold the 1930 panel and the 2030 panel up side by side as an interesting thought experiment if you saw someone this morning. The abstraction only survives at a distance.

And there is one more thing I would rather say plainly than imply. The future being described — agents triaging, routing, absorbing the simple cases so a physician's reach can stretch across a far larger population — is a future that runs on a great deal of software. He sells software in that category. That doesn't make him insincere and it doesn't make him wrong. But "we can't stop it" is a different sentence coming from someone positioned to build it, and it belongs on the table with everything else.

Having said that, I am not anti-software, and it would be dishonest to argue as though I were. I use Guava Health in my own practice (I've partnered with them on content, and I use them because they work), and it solves a problem I care enormously about: patient information arrives in silos — a scan here, a lab there, twelve years of history spread across four systems that don't speak to each other — and somebody has to turn that pile into a story. Software is extraordinarily good at that. Better than I am. Anyone who has tried to reconstruct a medical history out of a faxed stack of paper knows exactly how badly it's needed.

So I can appreciate what he is trying to solve without ever having used his product. We agree that fragmentation hurts patients. We agree that software ought to fix it.

Where we part is what you do with the result. There are two kinds of health technology. One of them hands the doctor a coherent story so that the thirty minutes in the room go deeper. The other uses the coherent story to argue that less of the doctor is needed. The first makes the relationship better. The second makes it optional.

Same tool. Opposite theory of what a doctor is for.

And the post did not leave that question open. It answered it.

The argument runs like this. Demand exceeds supply. The work has already been distributed once — to nurse practitioners, to pharmacists — and AI agents are simply the next step. Agents take the simplest cases, NPs and others take the middle, and the family doctor sees only what "truly requires a visit or oversight." Which means, in his words, that "we'll probably get to the point where a family doc doesn't know every patient all that well." And then the turn: "We can't stop it." So we should shape it instead of resisting it. "Protect the principles that truly matter, but be flexible on the details."

He is not cavalier about it. He says we will lose something. He sympathizes with family docs who see their profession "evolving from healer to software manager." He writes warmly about his own childhood doctor — the one who knew him, who gave him eye contact while charting five words on paper. I take all of that as sincere, and he is not wrong about the pressure building.

But sincerity isn't what I'm arguing with. It's the structure.

"We can't stop it" is not an observation. It's a rhetorical device, and an extremely effective one, because it converts a choice into a weather report. Once an outcome is inevitable, the only question left is how gracefully you accommodate it — and anyone still arguing about whether it should happen is just a sentimentalist standing in the rain insisting it isn't raining. Every genuinely contestable decision along the way — how many patients, how much of the visit, what gets triaged away and by whom — quietly gets reclassified as an implementation detail.

Which is precisely what that closing line does. "Protect the principles that truly matter, but be flexible on the details." It has the cadence of wisdom. But it never says which is which, and that omission is the entire move. He has just spent nine paragraphs sorting the relationship into "details" without once making the case that it belongs there.

I think it's the principle. And unlike the inevitability argument, I can show my work.

Every fiber of my being recognizes this as a dangerously misguided and untenable scenario, and I want to explain exactly why — starting with arithmetic, because the arithmetic alone should end the conversation.

The math

A typical primary care panel in this country runs somewhere between 2,500 and 3,500 patients. That is already too many. Ask any family doctor working in that system how well they know patient number 2,900.

In 2023, a team at the University of Chicago published a simulation in the Journal of General Internal Medicine asking a simple question: if a primary care physician actually delivered all the recommended preventive, chronic, and acute care to a panel of 2,500 patients, how long would that take? The answer was 26.7 hours a day.

Not 26.7 hours a week. A day. It breaks out to 14.1 hours of preventive care, 7.2 hours of chronic disease management, 2.2 hours of acute care, and 3.2 hours of documentation and inbox.

The same study then ran it again with a generous assumption — that the physician has a full team and delegates everything that can safely be delegated to nurses, pharmacists, educators, and staff. With a fully optimized team, the number comes down to 9.3 hours a day.

That is the best case, for 2,500 patients, with a whole team behind the doctor.

Now multiply the panel by four.

There is no AI agent that turns 26.7 hours into 24. There is no software that makes 10,000 relationships fit inside one human life. What software can do — what it is very good at doing — is generate the appearance of coverage. Messages get answered. Refills get approved. Boxes get checked. Quality metrics look outstanding. And the doctor whose name is on the panel has never laid eyes on the overwhelming majority of the people in it.

That isn't primary care. That's a call center with a medical license attached to it.

Ten thousand is a number you can put on a slide. It is not a number you can put in a waiting room.

The part that isn't about time

Here's what gets lost when we treat the panel as a throughput problem: the relationship isn't the packaging around the medicine. The relationship is the medicine.

A patient does not quit smoking because an AI agent sent a well-worded message about the risks of smoking. A patient quits smoking because someone they trust, who has known them for six years, who knows their mother died at 61, asks them about it for the fourth time in a way that finally lands.

A patient doesn't start a statin because a risk calculator fired. They start it because we talked through what the number actually means for them, and they decided.

A patient doesn't go get the colonoscopy because they got three automated reminders. They go because someone who knows they're terrified of the prep sat with that fear for a few minutes and helped them get over it.

Every one of those is an act of persuasion built entirely out of accumulated trust. You cannot automate accumulated trust. You can automate the reminder, and we have, and the numbers on preventive screening tell you how well that has worked.

And if that sounds like sentiment, it isn't. It is one of the better-evidenced findings in all of primary care.

A 2018 systematic review in BMJ Open examined 22 studies across nine countries with very different cultures and health systems. Eighteen of them — 82% — found statistically significant reductions in mortality associated with higher continuity of care. Not better satisfaction scores. Mortality.

The most striking single piece of evidence comes out of Norway, where researchers looked at 4.5 million people. Compared with patients who had been with their regular GP for one year, those who had been with the same GP for more than fifteen years had 25% lower mortality, 28% fewer acute hospital admissions, and 30% less use of out-of-hours care.

And it was dose-dependent. More years with the same doctor, better outcomes, in a steady line.

Sit with that, because it is the whole argument. The thing producing those numbers is duration of relationship. Not access. Not responsiveness. Not message turnaround time. The length and depth of one human relationship, measured in years.

A 10,000-patient panel runs that dial hard in the wrong direction. It is, functionally, a machine for reducing the number of patients a doctor can actually know.

Put it as plainly as it deserves: continuity lowers admissions. Panel size raises them.

And admissions, readmissions, ER utilization, chronic disease control are precisely the outcomes the entire industry is currently spending billions of dollars trying to move — the outcomes that continuity already moves, on its own, when you simply let it exist.

Creativity and humanity cannot be manufactured in vitro.

The house call was not telemedicine

There's a historical argument in the post that deserves a closer look, because it's carrying most of the weight — and it's the same argument the cartoon makes in pictures.

It goes like this. House calls were 40% of physician encounters in the 1930s, and under 1% by the 1980s. Today, 80% of office-based doctors do telemedicine visits — "the 2026 version of a house call." Go back and tell a doctor in 1980 that most of them would be making house calls again, and they'd never believe you.

It's a lovely symmetry. It also doesn't survive contact with the numbers.

Forty percent of encounters is not the same measurement as eighty percent of doctors. A physician who did one video visit all year counts in that 80%. To make the comparison honest you would need telemedicine's share of all encounters, which is nowhere near what house calls once were. The parallel holds up only because the denominator changed halfway through the sentence.

But set the arithmetic aside, because the deeper problem is that a video visit is not a house call in any sense that matters clinically. It is closer to being its opposite.

On a house call you find out the house smells like mildew, which might be why the asthma isn't controlled. You see how many stairs there are between the bedroom and the bathroom. You see what's in the fridge, and what's on the counter next to the pill bottles. You see whether there's one chair or two. You see the neighborhood.

On a video visit you get a carefully chosen four square feet, framed by whatever the patient wanted you to see. You learn what they meant for you to learn. And you cannot touch them, which matters more than the technology industry seems to believe, because a great deal of medicine still happens through the hands.

The house call put the doctor inside the patient's life. Telemedicine puts the patient inside a rectangle.

Those are not two versions of one thing, and calling the second the descendant of the first quietly discards the exact property that made the original worth having.

And look once more at the right-hand panel of that cartoon, because it gives the game away. There are a couple hundred houses in it, identical, marching off to the horizon — and there is not a single person visible in any of them. The people are off to the side in floating tiles, cut loose from the homes they live in. The patients have been separated from their houses. The doctor has been separated from both.

The caption says one doctor may come to 10,000 homes again. But nobody in that picture is in a home. They're in a frame.

So no — the 2030 panel is not an evolution of the 1930 panel. It's a different job.

What Chris Klomp said

At Hint Summit this past April, in a fireside chat I was sitting in the room for, Chris Klomp spoke about exactly this risk. Klomp is Chief Counselor of HHS and Director of the Center for Medicare, and previously served as Deputy Administrator of CMS. He is not a skeptic of technology — he ran a health tech company before he ran Medicare. And what he said was this:

"Give me a technology and I will show you how we destroy the value from it in healthcare. AI suffers that same risk."

He described the pattern of taking a genuinely valuable technology and "pile driving [the value] into the ground" — capturing it for the system rather than letting it reach the patient or the physician.

That is the conversation we should be having. Not how many patients one doctor's name can cover, but what happens to the time the technology gives back. There are only two answers. It returns to the exam room as attention, or it gets harvested as volume.

And this is exactly where the population and the person come apart. A dashboard can show 10,000 people better managed — screening rates up, refills current, messages answered inside the hour — while not one of those people has a doctor who knows their name. Both of those things fit on the same spreadsheet. Only one of them is care.

Because the part that keeps people alive, according to 4.5 million Norwegians, is the part no dashboard has a column for: whether anybody actually knows them.

A 10,000-patient panel is a proposal to harvest the time, and to call it innovation.

And here I want to concede something, because I think it matters.

Proposals like this one are going to keep coming. They will keep coming because the problem underneath them is real and the pressure is enormous, and they will often come from people whose jobs look nothing at all like mine — executives, engineers, analysts, investors, policymakers. That is not a bad thing. Those vantage points see things I cannot see from an exam room in Longmont, and a field with more people brainstorming solutions is healthier than a field with fewer. I don't want that to stop. I want more of it.

But the voice of primary care has to be in the room while it happens, and it has to be loud, and it has to keep saying the same unglamorous thing over and over: the relationship belongs at the center of care. That is not nostalgia, and it isn't a matter of taste. It is the variable with the outcome data attached to it.

That, more than any guardrail you could put on an agent, is what keeps us from pile driving the value of AI into the ground. Not better software. A primary care voice strong enough to insist that whatever the technology gives back belongs in the relationship.

The actual problem, and the actual fix

Let's be honest about why a 10,000-patient panel is even on the whiteboard. It's on the whiteboard because we are short of doctors and about to be shorter.

The AAMC projects a shortage of up to 86,000 physicians by 2036, including 20,200 to 40,400 in primary care specifically. Over that same window, the U.S. population grows about 8.4% — but the population over 65 grows 34.1%, and the population over 75 grows 54.7%. The people who need the most primary care are the fastest-growing group we have.

And the workforce is aging out from under us. 20% of active physicians are already 65 or older, with another 22% between 55 and 64 — meaning more than two of every five active physicians will be 65 or older within the next decade.

So yes. There is a real shortage, it's getting worse, and something has to give. On that, the post and I agree completely, and it's worth saying so plainly — because the shortage isn't the thing in dispute. It's the premise. It is the entire reason a 10,000-patient panel gets entertained as a serious idea in the first place.

Where we part is the conclusion drawn from it.

The answer to a shortage of doctors is more doctors. It is not spreading the ones we have thinner and buying software to hide the seams.

And here's the part the panel-size crowd keeps skipping: we know why primary care is unappealing. It isn't a mystery. Students and residents tell us directly. Three reasons come up over and over:

1. The volume. Twenty-plus patients a day, seven minutes of facetime with the doctor each, and the certain knowledge that you are doing a worse job than you were trained to do.

2. The administrative load. The documentation, the prior authorizations, the coding, and "pajama time" — the hours after dinner spent finishing charts while someone else puts your kids to bed.

3. The pay relative to everything else. Primary care sits at the bottom of the compensation table while carrying the broadest scope of any specialty.

Direct Primary Care fixes all three. Not partially. All three.

My panel is capped at a few hundred, not a few thousand. My visits are 30, 60, or 90 minutes because nobody is billing a code. There is no prior authorization department, no coding audit, no pajama time — I go home and I'm home. And the economics work, because the money goes from the patient to the doctor without passing through four buildings on the way.

Every doctor I know who has made this switch says a version of the same sentence: this is the job I thought I was signing up for.

And let me be clear that DPC is not the only road here. Those Norwegian numbers came out of a public, single-payer system built on personal lists — every citizen registered with a named GP. The mechanism was never the payment model. The mechanism is the relationship, and there is more than one way to build a system that protects it. Direct Primary Care is simply the version available to me, in this country, right now.

That's the fix. Not 10,000 patients. More doctors who want to stay, and panels small enough that staying means something.

Three asks

These are aimed at people training, practicing, or seeking care in the United States. Every system reaches continuity by a different road — Norway does it with personal lists inside single-payer — and the one I can speak to from the inside is mine.

If you're a medical student: go look at a DPC practice before you decide primary care isn't for you. Spend a day. What you were shown on your family medicine rotation was one version of this job, and it was the broken one.

If you're a resident: you do not have to take the hospital or corporate job. As of January 2026, 82% of American physicians are employed — 59.7% by hospitals and 22.3% by corporate entities — so I understand completely why it looks like the only door on the hallway. It isn't. The people who will tell you it's too risky to do anything else have generally never tried it.

If you're a patient: you are allowed to expect a doctor who knows your name without looking it up. If you don't have that, it's worth finding out what it costs to have it. For most people the answer surprises them.

That debate should not have been a debate. One doctor and 10,000 patients is not a bold vision of the future of primary care. It is the failure of primary care, running at scale, with better dashboards.

Enough is enough!

The post that prompted this is Dr. Joshua Liu's, and I've replied to him there as well. I have no quarrel with him personally, and I would rather argue with someone in public than about them in private. Chris Klomp's remarks are from the Hint Summit fireside chat in April 2026, which I attended; the quotations are my own transcription.

Dr. Brian Juan is a board-certified family physician and the founder of Metronome Family Medicine, a Direct Primary Care practice in Longmont, Colorado. Learn more at metronomemd.com or book a free consult at metronomemd.hint.com/booking.

sources:

Chris Klomp’s Fireside Chat at Hint Summit, April 2026

Dr. Josh Liu, MD’s original post one LinkedIn and his Website for SeamlessMD