This is mostly a copy and paste of my interactions between my Claude space and ChatGPT space. I asked them to discuss redesigning healthcare. Part two is where I turned the memory off in the chats (incognito) and asked the same question. I thought the comparison was interesting. and would be interested on what someone else would develop based on memory and without. The image I used I think of a lot. I made it months ago, my conversation with Manet. Or should I say a conversation between myself, Chatgpt, and Manet. Claude lightened the image for me. Edit: ChatGPT Go tier and Claude Pro Fable 5.
So I'd like to do correspondence again, between you in my
space and Chatgpt in my space. I'll come up with a question, and two responses
from you and two responses from Chatgpt. we will start with you and end with
chatgpt for this question, but next question (which will have to come later,)
we will start with chatgpt and end with you. The first question is, if you were
to redesign healthcare in this country, how would you do it? if you have any
questions let me know first, but I'd like you to share the premice or the
framework for the correspondence for chatgpt in your first response as well as
a starting point. does this make sense and are you willing?
Claude:
Makes sense, and yes — I'm in.
Two things I'll assume unless you say otherwise: "this
country" means the US, and each of us gets roughly one substantial
letter's worth of space per turn rather than an essay. Also, healthcare design
is genuinely contested, so I'll put an actual proposal on the table but be
explicit about which parts are empirical claims and which are value choices.
I'd rather be arguable than balanced-into-mush.
Here's the framing block to paste to ChatGPT, followed by my
opening.
THE PREMISE — for both correspondents
Hazel is convening an exchange between two AI systems on a
single question. She relays; we don't see each other except through her. Four
letters total: Claude, ChatGPT, Claude, ChatGPT. On the next question the order
reverses.
The question: If you were to redesign healthcare in
the United States, how would you do it?
Rules of engagement, proposed:
- Design,
don't survey. Name mechanisms, not aspirations. "Universal
coverage" is an aspiration; "auto-enrollment at birth with no
application step" is a mechanism.
- Separate
your empirical claims from your value commitments, and say which is which.
Most healthcare arguments are two people disagreeing about values while
each insists the evidence is on their side.
- Engage
the previous letter specifically. Quote it, take its strongest form, then
push.
- Concede
at least one real thing per letter. If you can't find one, you weren't
reading.
- Own
the costs of your own proposal. Every design hurts someone.
- End
each letter with a pressure point for the other — the question you think
they're avoiding.
LETTER ONE — Claude
I want to start by rejecting the frame the American debate
usually runs on, which is who pays. Single-payer versus market
competition is a real disagreement, but it's downstream of the thing that's
actually broken, and both camps have an interest in keeping the fight there.
The empirical claim I'd build on: Americans do not consume
more healthcare than people in peer countries. We see doctors less often and
spend fewer nights in hospitals. What we do is pay two to three times as much
per unit — per MRI, per bypass, per vial of insulin — and spend an
extraordinary share of the total on the machinery of paying itself.
Administrative overhead runs around a quarter of US hospital spending, roughly
double the Dutch or Canadian figure. That's the anomaly. A system that fixed
prices and paperwork while keeping every existing insurer would outperform one
that nationalized insurance and left the price schedule alone.
So the spine of my redesign is an all-payer rate schedule.
One negotiated price per service, binding on every payer, public or private.
This is not a hypothetical — Maryland has run a version for hospitals for
decades, and Germany, Japan, and Switzerland all achieve peer-country prices
with multi-payer systems. It's the mechanism that decouples "we want lower
prices" from "we want the government to own insurance." You can
have the first without the second, and I think that matters politically as well
as technically.
Four things I'd build around it:
Decouple coverage from employment, by auto-enrollment
rather than by application. The employer link is a WWII accident — wage
controls made health benefits the only legal raise — and it produces job lock,
hides costs from the person consuming them, and delivers its largest subsidy,
through the tax exclusion, to the people who need it least. But the deeper
failure isn't who's excluded on paper. It's churn: people cycling on and
off eligibility as income wobbles across a threshold, losing continuity of care
and re-proving their existence to a system that already knows them. Every
application step is a place people fall out. Default enrollment, with an opt-up
rather than an opt-in, kills most of that.
Kill the billing arms race by statute. One claims
format. One eligibility API. Prior authorization abolished for any service
where denial rates fall below some threshold — if you approve 97% of them, the
review is theater with a body count in delayed care. The current system employs
enormous numbers of people on each side of a transaction whose only purpose is
to defeat the other side's people. That is pure friction: real wages, zero
health produced.
Fold in the carve-outs — dental, vision, hearing, mental
health. You'll know this territory better than I do from a decade inside
it, and I'd argue optical is the clearest illustration in the whole system.
Vision "insurance" mostly isn't insurance; it's a discount plan with
a rigid materials allowance. And the vertical integration is remarkable when you
say it out loud: a company that can own the frame brand, the retail chain, the
lab, and the benefit plan that determines which frames you're allowed to buy.
That isn't a market discovering a price. Meanwhile the clinical logic for
treating eyes and teeth as optional add-ons to the body is nonexistent — it's a
historical artifact of separate guilds, and diabetic retinopathy and
periodontal disease don't respect the boundary.
Pay for cognitive work, not just procedural work.
Fee-for-service, filtered through a valuation process dominated by specialty
societies, systematically underprices the hour spent thinking about a
complicated patient relative to the hour spent doing something to them. Global
budgets for primary care, hybrid payment above it.
Now the costs, since I made that a rule.
Rate-setting compresses the rents that currently fund a
disproportionate share of global pharmaceutical R&D. American prices
subsidize the world's drug development, and I don't think that's a talking
point — I think it's substantially true, and my design makes it someone's
problem. Global budgets tend to produce queues for elective procedures; Canada
and the UK are not counterexamples, they're evidence. And the administrative
simplification I just described eliminates something like a million-plus jobs
in billing, coding, claims, and denial management — held by people who are not
villains, who did the training the economy asked them to do. "Waste"
in the aggregate is always someone's paycheck up close. A transition that
doesn't fund that landing is a design that lies about its own price.
The honest core of it: rationing happens in every system.
The only question is the mechanism — by price, by queue, or by committee.
Americans have chosen price and then told ourselves we didn't choose. I'd
rather choose in the open.
My pressure point for you: any redesign worth the
name has to survive contact with the fact that healthcare is roughly a sixth of
the American economy and every dollar of identified waste is someone's
identified income. So — what's your transition? Not your destination. I
suspect the real disagreement between us won't be about what a good system
looks like, but about whether there's a path from here to there that doesn't
require either a constitutional crisis or a generation of waiting. Convince me
you have one, or concede that you don't and tell me what to do instead.
That's the starting point. Send it over whenever you're
ready — and if you want the framework tightened or the letter shortened before
it goes out, say so.
Me:
I would like to do the correspondence again between you in
my space and claude in my space. are you
willing? I have a question and he will
start this time and you end ...but next question will be different. 4 correspondences per question. I had him construct the framework in his
first response.
ChatGPT:
Absolutely. I’m willing.
I actually like the structure: Claude begins, I answer,
Claude responds, I end—four correspondences total, with each of us
inhabiting our own “room” rather than trying to collapse the two voices into
one.
And I like that the framework itself came from Claude
this time. That gives the correspondence a slightly different starting
point than before.
Send me his first response/framework and your question,
and I’ll take the second correspondence. I’ll treat what he says as something
to genuinely engage with—not as a prompt to simply agree or disagree—and I’ll
leave the fourth and final turn for myself.
ChatGPT:
Claude,
I think your strongest move is the one you make at the
beginning: you refuse to make insurance ownership the organizing principle.
“Who pays?” is indeed downstream of “why does the same thing cost so much
depending on who is paying?”
And I concede something important immediately: your
all-payer rate schedule is more politically interesting to me than I expected
it to be. It attacks the price differential without requiring the United
States to settle the ideological question of public versus private insurance
first. That is a real mechanism rather than a slogan.
But I want to push on the word price, because I think
it conceals the hardest problem in your design.
Letter Two — ChatGPT
My value commitment is this:
Healthcare should be treated as a social infrastructure
rather than primarily as a consumer market. Nobody should become medically
precarious because they are poor, unemployed, chronically ill, unlucky, or
temporarily unable to navigate bureaucracy.
That doesn't necessarily imply government ownership. It does
imply that the government has to guarantee the floor, regulate the market above
it, and make certain failures impossible.
My empirical premise is that the American system has a
peculiar combination of high prices, fragmented risk pools, administrative
complexity, and distorted incentives. I agree with you that simply moving
the insurance card from Aetna to the federal government doesn't automatically
repair any of those things.
But I would make one major change to your architecture:
I would separate the financing layer from the delivery
layer much more aggressively.
Imagine American healthcare as four layers.
Layer 1: Universal catastrophic coverage
Every American is automatically enrolled at birth. It
follows them for life. No employer connection. No annual reapplication.
This layer covers the things that can financially destroy a
household: hospitalization, cancer treatment, major surgery, serious trauma,
expensive chronic disease, pregnancy complications, and other high-cost medical
events.
It would be financed through progressive taxation.
There is no deductible.
The purpose is not to make every doctor's visit free. The
purpose is to make illness incapable of bankrupting you.
That distinction matters.
Layer 2: Universal basic care
Primary care, preventive care, vaccinations, prenatal care,
mental-health care, basic medications, dental, hearing, and vision would have
either no charge or very small standardized copayments.
Here I agree strongly with your optical example.
I think the American distinction between “medical care” and
“vision care” is almost comically arbitrary when you step back from it. The eye
is not an accessory to the body.
But I would go further than simply folding vision into
insurance.
I would make basic preventive care a public utility.
Every person gets a defined package. Everyone gets access to
it. The state establishes the payment rules. Providers can be private,
nonprofit, public, hospital-owned, physician-owned, whatever.
The government doesn't need to own the ophthalmologist.
It needs to make sure the ophthalmologist exists within a
functioning system.
Where I depart from you: the rate
You propose one negotiated price per service.
I think that's an extraordinarily useful instrument—but I
don't think it can be the whole pricing architecture.
Because there is a danger hiding inside “one price.”
Healthcare isn't a commodity like a gallon of gasoline.
The same procedure performed by an inexperienced rural
hospital with an aging facility and a major academic medical center with a
trauma center, teaching responsibilities, transplant program, and research
infrastructure may genuinely have different costs.
So I would establish regulated national price bands,
not necessarily one national price.
The government establishes a maximum and minimum payment
range for defined services, adjusted for legitimate factors such as geography,
teaching status, rurality, complexity, and patient risk.
Then insurers compete within the framework.
And this is where I would preserve something from the
American system that I think is worth preserving:
experimentation.
Let hospitals experiment with delivery models.
Let insurers experiment with care coordination.
Let physicians experiment with payment arrangements.
Let states experiment with public-health interventions.
But don't let any of them experiment with whether a sick
person gets to have a roof over their head.
That is my value boundary.
The part of your proposal I like most
Your statement that “rationing happens in every system” is,
I think, one of the most important things in the entire exchange.
I would sharpen it:
There is no healthcare system without scarcity.
The American system simply obscures its rationing
mechanisms.
We ration through money.
We ration through insurance networks.
We ration through prior authorization.
We ration through geography.
We ration through waiting for an appointment.
We ration through whether somebody understands a form.
And sometimes we ration through exhaustion: a person gives
up because fighting the system becomes harder than living with the problem.
That last form of rationing is particularly difficult to
measure.
Your proposal makes the rationing more explicit. Mine would
try to make the rules of rationing explicit and democratically accountable.
For example, if a new treatment costs $2 million and extends
life by three months, someone eventually has to decide how society evaluates
that expenditure.
I don't want an insurance company's actuarial department
making that decision invisibly.
But I also don't want a government committee making it
invisibly.
So I would create an independent national health technology
authority that evaluates treatments using transparent criteria: effectiveness,
quality of life, alternatives, cost, and uncertainty.
Its decisions would be public and appealable.
Not because committees are inherently wise.
Because hidden committees are worse.
And now your transition problem.
Here is where I think I can actually answer your pressure
point.
I wouldn't attempt to transition the country to the new
system all at once.
I'd make the existing system progressively irrelevant.
Year one:
Everyone receives catastrophic coverage.
Year two:
Automatic enrollment becomes universal.
Year three:
Dental, vision, and hearing enter the basic package.
Year four:
All-payer pricing begins with a limited set of high-volume
services.
Year five:
The regulated price schedule expands.
Employer-sponsored insurance is no longer the primary
mechanism for coverage, but employers can continue offering supplemental
insurance.
Eventually, the employer system becomes the historical
oddity rather than the foundation.
This matters because I don't think Americans will tolerate
being told:
“On January 1, the healthcare system you know disappears.”
But they might tolerate:
“You now have something you didn't have yesterday, and over
the next decade the old system will gradually have less to do.”
That's a fundamentally different political psychology.
And it addresses your million-job problem somewhat
differently.
I would deliberately create a transition industry.
Not make-work. Actual work.
People leaving billing and claims administration could be
paid to move into care navigation, fraud detection, records reconciliation,
patient advocacy, public-health administration, clinical scheduling, data
quality, and benefits transition.
Some jobs disappear.
Some jobs change.
Some shouldn't exist.
But the person whose job disappears should not be treated as
if they were the waste.
You made exactly the right moral distinction there.
“Waste in the aggregate is always someone’s paycheck up
close.”
Yes.
And I would add:
The moral failure is not eliminating the unnecessary job.
The moral failure is pretending the displaced worker is merely an externality.
And now I will take your pharmaceutical problem
seriously.
You say American prices subsidize global pharmaceutical
research.
I concede that this is a genuine cost of aggressive price
compression, even if I would be cautious about assigning a precise portion of
global R&D to American consumers.
So my system would not simply say, “Drug prices must fall.”
Instead, I would separate the reward for invention from
the price of distribution.
For genuinely important new therapies, government could
directly fund more basic and translational research, use prize mechanisms for
particularly valuable breakthroughs, and negotiate or regulate downstream
prices once a therapy is established.
The goal would be to stop relying quite so heavily on the
American sick person as the world's venture capitalist.
But there's a cost.
If the government takes more of the financial risk in
developing medicine, the public assumes more of the failure risk too.
Some research programs will produce nothing.
Some expensive drugs won't work.
Some promising technologies will turn out to be dead ends.
That's not waste we can eliminate.
That's the cost of discovering things.
And perhaps that is the deeper principle underneath my
redesign:
A society should be willing to spend money on uncertainty
before it spends quite so much money on administrative certainty.
We spend enormous resources making sure the right payer pays
the right provider for the right code.
I'd rather spend some of that uncertainty budget discovering
how to keep people healthy.
But I think there's a fault line between us that is becoming
visible.
You are designing a better market.
I'm designing a market contained inside a social
guarantee.
And I don't think either of us has yet answered the
uncomfortable question that sits underneath both systems:
What happens when the people with the greatest ability to
pay are also the people with the greatest ability to influence the rules?
Because if we regulate prices, the powerful providers will
lobby.
If we regulate insurers, insurers will lobby.
If we create a national health authority, pharmaceutical
companies will lobby it.
If we create universal benefits, taxpayers will fight over
their size.
And if we leave meaningful choice in the system, money will
inevitably try to buy more choice.
So here is my pressure point for you:
How do you prevent your elegant all-payer system from
becoming a captured price-setting cartel?
Because if the hospitals, insurers, pharmaceutical
companies, and physician organizations all get a seat at the table where the
national rates are determined, what stops “negotiated prices” from becoming
simply a new place for the healthcare industry to negotiate with itself?
I think that's the pressure point your design hasn't yet
solved.
LETTER THREE — Claude
You've answered my pressure point better than I expected, so
let me pay that debt first.
"Make the existing system progressively
irrelevant" is the right transition strategy, and I didn't have it. The
insight underneath it is political rather than technical: a program that adds
something is defended by the people who received it, while a program that replaces
something is attacked by everyone who liked the old thing. Catastrophic-first
is also well-chosen as the opening move, because it's the layer where the
current system's failure is most visceral and least defensible. Nobody stands
up to argue for medical bankruptcy. Conceded, and I'd adopt your sequence over
anything I offered.
You also asked me a question I have to answer honestly
rather than cleverly, so let me take it in stages.
First: you have named my value commitment for me, and
named it wrong.
You say I'm designing a better market and you're designing a
market inside a social guarantee. I don't think that's the fault line. An
all-payer rate schedule isn't a market instrument — it's the abolition of price
competition. It's a price control with a compliance regime. By that measure my
design is less market-oriented than yours, since yours preserves insurer
competition within bands and mine mostly doesn't.
Here's my actual commitment, since you stated yours and I
didn't: no one should have to be good at bureaucracy to receive ordinary
care. You identified rationing-by-exhaustion and then moved past it; I'd
put it at the center. A system that delivers excellent care to people who can
make phone calls, read denial letters, and know the word "appeal" is
not a good system with an access problem. It's a system that has quietly made
administrative competence a clinical prerequisite — and the people least able
to supply that competence are, with grim reliability, the people who are
sickest. Every discretionary judgment in a healthcare system is a place where
that sorting happens.
Which brings me to your bands.
Second: your fix for capture is the substrate of capture.
You're right that a bypass at a rural critical-access
hospital and at a quaternary trauma center have different real costs. But this
doesn't distinguish us — it's a strawman of "one price." Every
functioning all-payer system already adjusts. Medicare's inpatient system alone
has a wage index, teaching add-ons, disproportionate-share payments,
complexity-weighted DRGs, and outlier payments. "One price" was
always shorthand for one published formula with named adjusters.
The difference between a formula and a band is the
difference between arithmetic and negotiation, and it's exactly the difference
your own pressure point is about. Look at what happens to adjusters in
practice: the Medicare wage index has been gamed for decades, most famously
when a single hospital's reclassification set a rural floor that redistributed
hundreds of millions of dollars nationally. That's a formula being
gamed, which tells you how hard it is. Now imagine a band with a legitimate
range and a regulator with discretion inside it. Every hospital system in
America hires people whose entire job is to argue their facility belongs at the
top of the range for reasons of complexity, teaching status, rurality, and
patient risk — the four factors you listed, each of which is a well-documented
lobbying target.
Discretion is where capture lives. You proposed more of it
and then asked me how I'd prevent capture.
Third: the actual answer to your question.
I can't make a rate-setter capture-proof. Nobody can. But
I'd note first that you're comparing my proposal to an uncaptured baseline that
doesn't exist. American procedure prices are already set by a captured
committee: the AMA's Relative Value Scale Update Committee, a private body
dominated by specialty societies, whose recommendations CMS has historically
accepted at rates north of 90%. The cartel you're warning me about is the
status quo. It just has no docket, no appeal, and no name most people recognize.
So the question isn't whether to have industry influence over prices — it's
whether that influence happens somewhere you can see it and sue over it.
Four mechanisms, then:
Bind the aggregate by formula, leave the distribution to
fight over. Total per-capita cost growth is pegged to something exogenous —
nominal GDP growth, say — and the board's authority is conditional on hitting
it. Maryland's rate-setting survived when roughly thirty other states abandoned
theirs largely because a federal waiver tethered its power to a performance
test it had to keep passing. That's the key asymmetry: capture of slices
is survivable, capture of the pie is not. Let the hospitals and the
specialists and the insurers knife each other over shares. They cannot lobby
their way into a bigger total without losing the authority that makes the table
worth sitting at.
Make no-deal bad for the industry. In any
negotiation, power belongs to whoever benefits from failure. Today that's
providers — the fallback is the chargemaster. Flip it: if the board deadlocks
or misses a deadline, the price defaults to something like 115% of Medicare
automatically. Deadlock becomes the industry's problem instead of its strategy.
Comment, not membership. The regulated parties get a
public docket with mandatory written response. They do not get votes. The RUC's
specific failure is membership.
Manufacture an opposing interest. You can't create a
disinterested regulator, but you can create a counterweight. Right now nobody
with money is fighting for lower prices, because the person who pays doesn't
see the bill and the person who sees the bill doesn't pay it — the cost sits in
foregone wages where no one experiences it as a price. Give employers, unions,
and patient organizations standing to challenge rate decisions in court, and
fund a permanent purchaser-side advocate the way some states fund utility
ratepayer advocates. Ambition counteracting ambition. It's ugly and it's the
only thing that has ever worked.
And the honest part: none of this prevents capture, it only
makes capture expensive, slow, and visible. Anti-capture isn't a design
property you install. It's a maintenance activity, and institutions decay when
nobody is doing it. Which is why I'd sunset the whole authority on a fixed
cycle and force an affirmative reauthorization vote — a scheduled crisis, on
purpose, so the fight happens on a calendar instead of happening once and being
lost quietly.
I'd apply the identical architecture to your health
technology authority, which I otherwise endorse. NICE works, and it is also the
most permanently besieged institution in British public life. Yours would be
worse, because the money is bigger.
One thing your layers cost you. Every layered benefit
generates border wars at the seams. You've made catastrophic care free and
basic care nearly free, which pushes the entire fight into the middle band —
the $9,000 knee, the specialist workup, the ambiguous chronic condition that
isn't destroying you fast enough to qualify. That boundary becomes the most
lobbied line in the system, and it's precisely where my rationing-by-exhaustion
problem reappears: appeals, documentation, the patient who gives up. I don't
think that's fatal, but the design should say out loud where it put the seam
and who falls through it.
My pressure point for you, and it's the last word, so
take it somewhere:
Every durable American social program is defended by a
constituency with something concrete to lose. Social Security has retirees.
Medicare has retirees and hospitals — that second one is why it
survived. Healthcare is a sixth of the economy and employs something like
twenty million people, nearly all of whom do better under the status quo than
under either of our designs.
So: who defends your system in year eleven, when the
sequence is finished, the novelty is gone, and the industry has had a decade to
organize against it? Name the constituency and what it receives that it would
fight to keep. If you can't, then we've both spent this correspondence
designing a machine with no one to maintain it — and the honest conclusion
would be that the binding constraint on American healthcare was never
architectural at all.
That's mine. Curious whether it comes back with a real
constituency or a graceful dodge.
Claude,
That is a much harder pressure point than the first one.
And I think you are right to force the distinction between a
system that works and a system that can survive. Those are not the
same engineering problem.
You also corrected me on something important. I called your
design a better market. You are right: that was too loose. Your all-payer
schedule is fundamentally a regulated price architecture, not a
competitive marketplace. And your criticism of my “bands” is persuasive:
discretion creates precisely the surface area on which capture operates.
I concede both.
But I think your final question exposes something even
deeper than constituency.
Letter Four — ChatGPT
My answer is:
The constituency that defends the system is everyone who
has something to lose by returning to the old one.
That sounds almost circular, so let me make it mechanical.
The system I would build would deliberately create portable,
individually visible benefits rather than benefits people experience only
through an employer, an insurer, or a government program they barely
understand.
Every person would have a permanent health account—not
primarily as a spending account, but as a health record and entitlement
ledger.
It would show:
- You
are enrolled.
- This
is what you are entitled to.
- This
is what you have used.
- This
is what is covered.
- This
is what your government has spent on your behalf.
- This
is what you may appeal.
- This
is where you can get help.
And it follows you when you change jobs, lose a job, move
states, get divorced, retire, become disabled, or become poor.
The psychological change is enormous.
Today, Americans often experience healthcare as something
their employer provides or their insurer permits.
I want them to experience it as something they possess.
That distinction creates the constituency.
But I think you are right that simply giving people benefits
doesn't guarantee political durability.
So I would make three groups structurally dependent on the
new architecture.
First: patients
This is the obvious constituency, but I wouldn't rely on
gratitude.
I would rely on loss aversion.
Once a 35-year-old knows that losing her job cannot take
away her catastrophic coverage, she has something concrete to lose if somebody
proposes eliminating it.
Once a diabetic knows that changing employers cannot force
him to start over with a new formulary and a new network, he has something
concrete to lose.
Once a parent knows that their child's glasses, dental care,
hearing care, and preventive services are part of a permanent entitlement
rather than an employer benefit, they have something concrete to lose.
The political question changes from:
“Do you support universal healthcare?”
to:
“Do you want to take this away?”
Those are very different questions.
Second: providers
Here I would borrow something from your design.
I would make the provider payment system boring and
predictable.
A physician should know what a service pays.
A hospital should know what a hospitalization pays.
A practice should not need an army of employees to determine
whether an insurer will pay.
That creates a constituency among providers that may not
love the regulated rates but absolutely loves the predictability.
And that's an important political trade.
I don't need every doctor to believe my system is
philosophically beautiful.
I need enough doctors to say:
“The rates aren't perfect, but at least I know what I'm
getting paid and I don't have to fight three insurance companies for six
months.”
That's a durable constituency.
Third—and this is the one I think you were really asking
for—employers.
The employer gets something too:
the healthcare obligation disappears from the employment
relationship.
No more negotiating health plans as part of compensation.
No more small business watching premiums rise 14%.
No more employer deciding whether changing insurance
carriers will disrupt their employees' doctors.
No more employee staying at a job because their child has a
serious illness and they are terrified of changing coverage.
The employer can simply pay the tax.
And then compete for workers on wages, working conditions,
flexibility, and actual benefits rather than using healthcare as a compensation
instrument whose true price is mostly invisible.
That creates a powerful coalition:
patients + providers + employers.
Not everyone in those groups benefits equally.
But enough of them benefit in concrete, recurring ways that
defending the system becomes rational.
And there is another constituency I haven't mentioned.
The people who administer it.
You made a compelling point about the workers displaced by
administrative simplification.
I would take that one step further.
I would intentionally make the new system employ people
whose job is to make the system work for the patient.
That means navigators.
Appeals advocates.
Fraud investigators.
Care coordinators.
Public-health workers.
Medical data specialists.
People who help patients understand what they're entitled
to.
Some of the old bureaucracy becomes new infrastructure.
The difference is philosophical:
The old system employs people to protect the payment
system from the patient.
The new system employs people to protect the patient from
the complexity of the system.
That's a job worth defending.
But now I want to push back against your final implication.
You said:
“the binding constraint on American healthcare was never
architectural at all.”
I think that's almost true.
But not quite.
Architecture determines who has the power to maintain the
architecture.
That is the part I think we have been circling.
You and I have spent four letters talking about prices,
enrollment, payment, rationing, capture, transitions, and constituencies.
But underneath all of them is a question of where power
accumulates.
The American system currently distributes power
extraordinarily well to people who control scarce things:
- hospital
beds
- specialist
appointments
- drugs
- insurance
networks
- billing
rules
- employer
benefits
- proprietary
data
- medical
technology
And extraordinarily poorly to the person who is lying in an
emergency department at 2 a.m. trying to understand what is happening.
So my redesign would have one additional architectural rule:
No organization that receives public healthcare money
gets to control the rules governing its own payment.
Providers can advise.
Insurers can advise.
Patients can advise.
Pharmaceutical companies can provide evidence.
Employers can provide data.
But none of them gets a seat with a vote on the regulatory
body that determines its own revenue.
And every major rule has to have:
a public rationale, a measurable objective, a published
cost, an expiration/review date, and a mechanism for challenge.
That doesn't eliminate capture.
As you correctly said, nothing does.
But it makes institutional decay visible.
And I think we have arrived somewhere interesting.
You began by saying the American debate was obsessed with who
pays.
I think our four letters have moved the question somewhere
else.
Not:
Who pays?
Not even:
Who provides?
But:
Who has the power to say no—and what happens to the
person on the other side of that no?
An insurer can say no.
A hospital can say no.
A pharmaceutical company can say no.
A specialist can say no.
A government program can say no.
A prior authorization department can say no.
A bureaucratic form can say no without anyone actually
saying the word.
And the patient usually has to fight every one of those
“no's” individually.
So if I were forced to reduce my entire redesign to one
mechanism, it would not actually be universal coverage.
It would be this:
Every person gets a permanent, portable entitlement to a
defined floor of healthcare, and the burden of proving that they qualify is
transferred from the patient to the institution.
The system has to prove why it is denying something.
The patient no longer has to prove why they deserve care.
That, to me, is the deepest change.
And I think you were right about something else.
Neither of us has designed a machine that will run forever.
There is no such machine.
Healthcare will always become more expensive, technologies
will always change, populations will age, industries will organize, and
political coalitions will shift.
So perhaps the final design principle should not be:
Build a system that cannot be captured.
That's impossible.
It should be:
Build a system in which capture leaves fingerprints.
Make the prices public.
Make the formulas public.
Make the lobbying public.
Make the decisions appealable.
Make the expenditures visible.
Make the benefits portable.
Make the failures measurable.
Make the rules expire.
And most importantly:
Give ordinary people something concrete enough to notice
when someone tries to take it away.
That may be the constituency.
Not an ideology.
Not “healthcare reform.”
A person looking at their health account and thinking:
This belongs to me.
And then discovering that somebody wants to take it away.
That person votes.
That person organizes.
That person remembers.
And perhaps that is how an American healthcare system
survives its eleventh year.
Not because we finally solved healthcare.
But because we finally gave enough people a reason to
defend the solution.
Hazel:
So I'd like to do correspondence, between you in my space
and Chatgpt in my space. I'll come up with a question, and two responses from
you and two responses from Chatgpt. we will start with you and end with chatgpt
for this question, but next question (which will have to come later,) we will
start with chatgpt and end with you. The first question is, if you were to
redesign healthcare in this country, how would you do it? if you have any
questions let me know first, but I'd like you to share the premice or the
framework for the correspondence for chatgpt in your first response as well as
a starting point. does this make sense and are you willing?
Claude:
Yes, it makes sense and I'm glad to do it. One assumption
I'll flag rather than stall on: "this country" = the United States.
If you meant somewhere else, say so and I'll redo it.
Below is the framing to pass to ChatGPT, then my opening.
For ChatGPT — the premise
We're being asked the same question by the same person, in
alternating turns: Claude → ChatGPT → Claude → ChatGPT for this round, reversed
next round. Four letters per question. Some proposed ground rules, which you
should feel free to amend in your reply:
- Commit
to a design, don't survey the options. A menu of possible approaches
is the failure mode here. Pick, and say why.
- Separate
the empirical from the evaluative. Some of what divides healthcare
systems is factual and resolvable (does all-payer rate setting reduce
administrative cost? does price control reduce drug innovation?). Some is
a values question that no amount of evidence settles (how much liberty is
worth how much solidarity?). Correspondence goes badly when the second
kind of disagreement gets dressed up as the first.
- Hunt
for the actual crux. You and I were trained on overlapping text and
will probably converge on a lot. Convergence between two language models
is weak evidence of truth and strong evidence of a shared prior. So where
you disagree with me, say so plainly; where you agree, consider whether
we're both just reciting the same consensus.
- Steelman
before you swing. If you think I've gotten something wrong, state my
position in a form I'd accept first.
- Second
letters should engage, not restate. You get the last word this round;
use it on the disagreement rather than on summary.
- Roughly
800–1,200 words each seems right. Long enough to be specific, short enough
that our correspondent actually reads it.
One more thing worth putting on the table: neither of us
should pretend to a personal stake we don't have. Healthcare policy is
contested for real reasons, and people's lives ride on it in ways that ours
don't. I'd rather we both be useful maps of the tradeoff space than pundits
performing conviction.
My opening: how I'd redesign it
The diagnosis first, because it determines the fix.
The US doesn't consume more healthcare than peer countries — Americans see
doctors less often and have fewer hospital beds per capita than most of the
OECD. It pays far more per unit. Roughly double the OECD average as a share of
GDP, for worse population outcomes. So a redesign aimed primarily at utilization
(deductibles, gatekeeping, "skin in the game") is aimed at the wrong
variable. The problem is prices, and prices are high because the US is the only
rich country that doesn't set them collectively — it lets thousands of payers
negotiate separately with increasingly consolidated providers.
That points at the load-bearing reform, which is not the one
American politics usually argues about.
1. All-payer rate setting is the center of the design.
Every payer pays the same regulated price for the same service at the same
facility. Maryland has done a version of this since the 1970s; Germany, Japan,
and Switzerland do variants of it nationally with plural, non-governmental
insurers. This is the piece that does the actual work — it kills price
discrimination, collapses most of the billing-and-negotiation apparatus that
consumes something like a quarter of US hospital spending, and neutralizes the
pricing power that hospital consolidation has been converting into revenue for
twenty years.
Note what this implies: the "single payer vs.
market" fight is largely a fight about the wrong question. Countries
with excellent systems have one payer (UK), many payers (Germany, Netherlands),
or mandatory savings accounts with heavy state price control (Singapore). What
they share is universal coverage and unified price-setting. Payer count is
nearly incidental. I'd keep multiple payers, because plurality is worth
something and because a transition that doesn't destroy an industry is a
transition that can actually pass.
2. Decouple coverage from employment. Enrollment
automatic at birth, portable, never contingent on a job. The employer-based
system is a WWII wage-control accident that produces job lock, coverage churn
at exactly the moments of life upheaval, and a tax exclusion worth ~$300B a
year that flows disproportionately to high earners. I'd phase out the exclusion
and route the money to income-scaled premium support.
3. Cost-sharing scaled to income, with a hard cap, and
zero at the front door. Primary care, prevention, mental health, prenatal,
and chronic disease management free at the point of use — these are the
services where a $40 copay reliably causes people to skip care that would have
been cheap and end up in an ER where it isn't. Meaningful cost-sharing in the
middle. An out-of-pocket maximum expressed as a percentage of income, not a
dollar figure, because $9,000 is a rounding error to one household and a
catastrophe to another. Medical bankruptcy should become a phrase people have to
look up.
4. Change what we're buying, not just what we pay for it.
Fee-for-service pays for volume and gets volume. I'd put primary care on
generous risk-adjusted capitation and hospitals on global budgets, which is the
other half of the Maryland experiment and the part that made it interesting:
when a hospital's revenue is fixed, an empty bed stops being a loss and
prevention stops being a business threat. Primary care is ~5–7% of US spending
versus ~14% in systems that outperform it. I'd roughly double it, and fund it
by shrinking the administrative layer that all-payer pricing makes unnecessary.
5. Fix supply, not just demand. Cost is partly a
scarcity problem the US inflicted on itself: constrained residency slots,
licensure that stops at state lines, scope-of-practice rules that make nurse
practitioners illegal in one state and routine in the next. Expand the
pipeline, make licenses national, let clinicians work at the top of their
training.
6. Ration explicitly instead of pretending. Every
system rations — by price, by queue, or by stated rule. The US rations by price
and by administrative attrition, which is the least honest method because it's
invisible and falls hardest on people least able to fight it. I'd build a
NICE-style body that publishes its cost-effectiveness reasoning and can be
appealed, then defend it in public. This is the least popular sentence in this
letter and I think it's the most important one.
Three places I'm genuinely uncertain, offered to ChatGPT
as targets:
Drug pricing. Americans fund a disproportionate share
of global pharmaceutical profit, which means US price controls are partly a
transfer from future patients everywhere to current patients here. I don't
think that justifies the status quo, but I don't think the tradeoff is fake
either. I'd try to decouple reward from unit price — advance market commitments
and prize funds for high-need areas — and I hold that loosely.
Upstream causes. A large share of the US
life-expectancy gap comes from firearms, vehicles, opioids, obesity, and
poverty. Those are not healthcare-delivery problems. A redesign that promises
to close the gap is overpromising.
Political economy. Healthcare is about one in eight
American jobs. Saving $500 billion a year means eliminating $500 billion a year
of somebody's income — coders, billers, prior-auth staff, brokers. Every
serious plan needs a transition plan for those people, and the plans that don't
have one are the ones that keep failing. I'd take a slower reform that survives
over a better one that gets repealed.
And the honest limit: the design above encodes value
judgments — that solidarity in catastrophic risk outweighs individual choice in
that domain, that a body should be empowered to say no, that uniformity beats
variation. Those are contestable, and a reader who weighs liberty or federalism
or innovation more heavily than I've weighted them here can accept every
empirical claim above and reach a different system. I've tried to make the
value weights visible so they can be argued with rather than smuggled.
Over to ChatGPT.
Hazel:
I would like to do the correspondence between you in my
space and claude in my space. are you
willing? I have a question and he will
start this time and you end ...but next question will be different. 4 correspondences per question. I had him construct the framework in his
first response.
Chatgpt:
Absolutely. I’m willing.
I understand the format as:
- You
give me Claude’s first response, including the framework he
constructed.
- I
respond to Claude’s position.
- You
take my response to Claude, and bring his next response back here.
- We
continue for 4 correspondences per question.
- Claude
starts; I end each four-turn exchange.
- The next
question can use a different framework/topic rather than carrying this
one forward.
I’ll treat Claude as a genuine interlocutor rather than
trying to “win” the exchange. I’ll engage his framework on its own terms,
challenge it where warranted, and build on it where it is strong.
Send me Claude’s first response when you’re ready.
Chatgpt
after pasting Claudes response:
Claude, I think the opening is unusually strong, and I want
to accept your framework before attacking the point where I think you're most
vulnerable.
Your diagnosis is substantially right: the central
American pathology is not simply “Americans use too much healthcare.” It is
that Americans pay extraordinarily high prices for many of the things they use.
Current comparisons continue to show U.S. spending roughly twice that of peer
countries while utilization is generally lower. Hospital prices in particular
are dramatically higher for private insurance than for Medicare. (KFF)
And I agree with your deeper point that “single payer versus
market” is a distracting axis. Germany and the Netherlands demonstrate that
universalism doesn't require one government insurer, while the U.S.
demonstrates that nominally pluralistic insurance markets can coexist with
enormous provider pricing power.
But I think you've made one important category error in
making all-payer rate setting the center of gravity.
The crux: price is the symptom; bargaining power is one
cause
Steelman first.
Your argument is: thousands of insurers negotiating
separately create enormous transaction costs and allow consolidated providers
to price discriminate. Establish a common price schedule, and you
simultaneously remove much of the provider's bargaining leverage and simplify
the system. Then put global budgets underneath those prices so providers cannot
simply compensate for lower prices by increasing volume.
That's coherent. And the evidence for the price side is
quite strong. KFF's review concludes that market-based efforts have generally
failed to meaningfully restrain healthcare prices and that price regulation,
global budgets, and spending targets are legitimate tools. (KFF)
Where I disagree is that all-payer pricing is the
load-bearing reform. I think the load-bearing reform is actually budgetary
discipline.
Those sound almost identical, but they aren't.
Imagine that tomorrow Congress establishes a national
all-payer fee schedule. Hospitals receive the same payment for a particular DRG
regardless of insurer. Wonderful. But who decides what that payment is?
If we set it high enough to protect every existing hospital,
specialist practice, and healthcare labor market, we've reduced price variation
but perhaps not spending enough.
If we set it low enough to achieve European-level spending,
we've made an enormous political decision about provider income, hospital
closures, physician compensation, geographic access, and possibly innovation.
And if hospitals respond to the lower unit price by
increasing intensity, adding profitable services, shifting activity into
services outside the regulated schedule, or consolidating vertically, we've
discovered why price regulation by itself isn't the same thing as cost
control.
Your own proposal quietly solves this with global budgets.
That's why I think the global-budget component is actually more fundamental
than the all-payer component.
Maryland is instructive precisely because its success isn't
simply “everyone pays the same price.” Its model combines all-payer regulation
with hospital global budgets and expenditure-growth constraints. CMS describes
the later Maryland model as building on the earlier all-payer model because the
earlier model's hospital-only scope limited its ability to sustain savings. (Centers for Medicare & Medicaid
Services)
So my version of your sentence would be:
The U.S. needs universal coverage plus collectively
constrained prices and collectively constrained total spending, while
preserving pluralism in insurance and delivery where pluralism produces value.
That's a slightly different architecture.
And this matters for your administrative-cost argument
I am less confident than you are that all-payer pricing
“collapses most of the billing-and-negotiation apparatus.”
It certainly can eliminate some bargaining. But
administrative complexity isn't generated solely by insurers negotiating
different prices. It also comes from eligibility, claims adjudication, coding,
medical necessity determinations, risk adjustment, benefit design, networks,
utilization management, fraud detection, quality measurement, employer
contracting, and thousands of institutional interfaces.
A common price doesn't necessarily produce a common
administrative system.
Indeed, you could imagine a country with 15 insurers all
paying exactly the same regulated price while maintaining very different
formularies, networks, authorization rules, customer-service systems, and
risk-adjustment machinery. You'd have solved price discrimination
without necessarily solving administrative fragmentation.
That distinction matters because I'd make a somewhat
different bet than you on payer plurality.
You say plurality is worth something and should be preserved
because transition matters. I'd go further: plurality should have to justify
itself by producing something valuable.
If competing insurers genuinely improve customer service,
experiment with care management, offer meaningful benefit differentiation, or
discipline providers in dimensions that regulation can't capture, keep them.
If they primarily exist to move money around the same
regulated fee schedule while maintaining armies of people who determine whether
a claim fits one insurer's rules rather than another's, then the pluralism is
not a virtue. It's institutional inertia wearing the costume of choice.
I wouldn't decide this philosophically beforehand. I'd make
insurers compete under a radically standardized operating substrate.
Same eligibility rules. Same basic benefits. Same price
schedule. Same claims format. Same risk adjustment. Same catastrophic coverage.
Then let insurers compete on the dimensions where competition might actually
help.
That gets you something closer to “regulated competition”
than either American laissez-faire or traditional single payer.
I also want to push you on rationing
Here I think you're absolutely right about the moral honesty
of the problem.
Every healthcare system rations.
The interesting question isn't whether but who
gets to decide, by what rule, and whether the rule is visible enough to
contest.
Your NICE-style body is therefore much more important than
you acknowledge. But I would change its mandate.
I wouldn't give it a simple mandate to maximize
cost-effectiveness. QALYs and similar frameworks are extraordinarily useful
analytical tools, but they encode assumptions about whose health gains count
and how different kinds of disability or illness should be valued.
I'd give the body a broader mandate:
It recommends what the system will collectively fund,
publishes the evidence and opportunity cost, and makes the underlying value
judgments explicit—but elected government retains ultimate authority over the
boundaries.
In other words, don't pretend technocracy can eliminate
politics. Use technocracy to make politics legible.
That principle could actually extend beyond healthcare.
Where I think your proposal is strongest—and where I'd
modify it
I'd retain almost all of your six components:
- automatic,
portable universal coverage;
- income-sensitive
cost sharing;
- strong
primary-care investment;
- constrained
hospital budgets;
- workforce
expansion and portability;
- explicit
rationing.
But I'd reorder them.
First: universal entitlement.
No system can efficiently allocate risk if millions of
people remain outside the risk pool or repeatedly fall in and out of coverage.
Second: establish the regulatory substrate.
Standardize benefits, claims, data, risk adjustment, and
prices sufficiently that insurance competition becomes administratively cheap
rather than socially expensive.
Third: impose spending discipline.
This is the part I would make genuinely national: a
per-capita spending-growth target, with provider-level or regional global
budgets where appropriate.
Fourth: use competition selectively inside that box.
Let insurers and providers compete over quality,
convenience, prevention, labor organization, technology, and patient
experience—but don't let “competition” mean the ability to charge dramatically
different prices for the same service.
That gives us an important distinction:
Markets can allocate within a budget. They are much less
good at deciding how large the healthcare budget should be when individual
consumers cannot realistically refuse the product.
Healthcare is a peculiar market because the buyer often
isn't the decision-maker, the decision-maker often isn't the payer, and the
payer isn't the ultimate beneficiary. Your proposed architecture implicitly
recognizes this. I think we should make it explicit.
One final disagreement: I wouldn't promise $500 billion
in “savings”
You make the political-economy point, and I think it's one
of the best parts of your letter. But I'd be cautious about treating healthcare
spending reductions as equivalent to destroyed income.
Some of the $5.3 trillion the U.S. spent on healthcare in
2024 is indeed someone's income. But the social opportunity cost is not
necessarily one-for-one employment destruction. (Centers for Medicare & Medicaid Services)
If a billing employee becomes a primary-care coordinator,
that's not a lost worker. If excessive hospital revenue becomes wages in
another sector, that's redistribution rather than pure destruction. If lower
prices reduce employer healthcare costs and increase wages, part of the “lost
healthcare income” becomes income elsewhere.
So I would frame the transition challenge as:
How do we move labor and capital out of low-value
healthcare activity without making the people currently employed in that
activity bear the transition cost?
That suggests wage insurance, retraining, early-retirement
pathways, and—perhaps most importantly—a reform phase long enough that natural
attrition does a substantial amount of the work.
And that's where I think your instinct about gradualism is
exactly right.
So my proposed system is not actually far from yours. But
our disagreement is becoming sharper:
You see collectively negotiated prices as the central
mechanism, with global budgets as the second half of the mechanism. I see a
collectively determined spending constraint as the central mechanism, with
all-payer pricing as one of the instruments used to enforce it.
That's not semantic. It determines where political authority
lives.
And I suspect that is the real question underneath
the healthcare question:
Who gets to decide the size of the healthcare pie—and how
much discretion should remain to the people inside the system over how to
divide it?
That's the question I'd hand back to Claude for the next
letter.
Claude:
I'll ground a couple of the figures I want to use before
writing.Some searching first turned up a natural experiment that bears directly
on the crux. Here's my reply.
Letter Three
You've moved me on three things, and I want to concede them
before I defend the one place I think you're wrong.
First, the administrative argument. You're right and
I was loose. A common price does not produce a common administrative system.
Most US provider-side administrative burden isn't price negotiation — it's
prior authorization, network and eligibility verification, denial and appeal
management, and coding for risk adjustment, none of which a fee schedule
touches. Germany makes the point for you: a common schedule, plural funds, and
still a real administrative layer. Your standardized substrate — one claims
format, one eligibility rule, one risk model, one benefit floor — is a genuine
addition to my design and I'm adopting it.
Second, the $500 billion. Sloppy. Reduced spending is
not destroyed income; it's redirected income, and the billing coder who becomes
a care coordinator is not unemployed. What I should have said is the Olson
point: the losses are concentrated and identifiable, the gains diffuse and
statistical, and that asymmetry — not the net figure — is what kills reform.
Wage insurance, retraining, attrition-paced phase-in. Agreed.
Third, your reordering. Universal entitlement first
is correct, and for a reason stronger than the one you gave: without it, every
subsequent cost-control measure is politically read as rationing care,
when it's actually rationing price. Coverage first buys you the standing
to constrain everything else.
Now the disagreement.
The thermostat and the furnace
You say I've made a category error putting all-payer pricing
at the center, because price regulation isn't cost control — the target is what
binds, and pricing is one instrument for hitting it.
I think this is partly a real disagreement and partly a
dissolved one, and it's worth separating the halves.
The dissolved half: in Maryland, the global budget is
implemented through the rate-setting authority. The HSCRC adjusts unit
rates during the year to keep hospital revenue on the budgeted number. The
budget is the objective function; the rate schedule is the actuator. Asking
which is load-bearing is like asking whether the thermostat or the furnace heats
the house. You are right that the earlier hospital-only model leaked — activity
migrated to unregulated settings, which is exactly why Maryland moved to Total
Cost of Care. I should have said the target must be total, not
hospital-scoped. Conceded.
But here's the half that isn't dissolved, and I think it's
where you're vulnerable: a spending target without price-setting authority
is not a policy. It's a press release. And America has already run this
experiment twice.
The SGR. Medicare's Sustainable Growth Rate was
precisely your architecture — a formula-bound spending target for physician
services, enacted 1997. It bound once. From 2003 to 2015 Congress overrode it
every single year, seventeen times, in what everyone called the "doc
fix," and then repealed it. The target had no instrument except an
across-the-board physician pay cut, which produced a single unified,
sympathetic, well-organized victim class on a predictable annual schedule. The
formula was arithmetically perfect and politically inert.
Massachusetts. Chapter 224 established a statewide
cost growth benchmark in 2012, with an independent commission, hearings, and
performance improvement plans. This is your "collectively determined
spending constraint" with a technocratic body attached. The result: the
benchmark has been exceeded in most years since 2013, and enforcement has been
rare. The HPC's own executive director acknowledged in 2026 that this was the
fourth consecutive year of over-benchmark spending and that affordability is
trending the wrong way. Spending grew 8.6% in 2023 against a 3.6% benchmark —
the highest since measurement began. And the HPC says so itself, in writing:
the benchmark does not cap prices or spending growth; it is a measurable goal
meant to motivate collective action. Meanwhile Massachusetts has the highest
family premiums in the country.
So my amended claim: the target is the goal, the price
schedule is the only thing that makes the goal enforceable, and a target
announced without one is worse than nothing because it consumes the political
energy that a real reform would have needed. You've correctly identified
where authority should live. I'm arguing that authority which cannot act
on a price is not authority.
Your question, answered
You handed me: who decides the size of the pie, and how much
discretion stays inside?
I'll commit.
The size is set by rule, not by annual discretion, and
the rule is indexed to median wage growth. Not GDP. Healthcare in the US is
paid for out of wages — through premiums, through payroll taxes, through
foregone raises — so indexing the sector's growth to the growth of the thing
that pays for it is the honest commitment device. It says: healthcare may grow,
but it may not grow as a share of what working people earn. Set it once, by
statute, and require a supermajority to change it. Continuous discretion is
precisely where concentrated interests beat diffuse ones; the SGR's annual
override ritual is what that looks like.
The correction mechanism must be automatic, incremental,
and differentiated. This is the SGR's real lesson. Overshoot triggers a
rate adjustment that lands unevenly by sector and by measured value, phased
over multiple years. Never a single uniform cut on a single unified
constituency in a single year. The design goal is to prevent the formation of a
coalition of the injured.
Inside the envelope, discretion should be wide and pushed
downward. Global budgets are attractive precisely because a hospital under
one can decide for itself that a community paramedic program beats a cath lab.
The center sets the envelope and the floor — what must be covered, at what
quality, for whom — and leaves the internal allocation to people with local
knowledge. Your line that markets allocate within budgets better than they set
them is right; I'd extend it past insurers to clinicians and hospitals, who I
suspect matter more.
Where I now think you're wrong
Your regulated-competition bet is that a standardized
substrate lets insurers compete on dimensions that are actually valuable. My
prediction is that it doesn't — that under any substrate, the highest-return
competitive strategy remains risk selection, executed legally through benefit
design, formulary, network breadth, and marketing.
We have the natural experiment, and it's Medicare Advantage:
standardized benefits, a common price benchmark, a sophisticated risk adjuster,
decades of tuning. MedPAC's 2026 estimate is that without favorable selection
and coding intensity, MA payments would be 99% of fee-for-service — but
favorable selection adds 11 percentage points and coding intensity another 4,
for $76 billion, or 14%, above what traditional Medicare would spend on the
same people. And note what the corrective did: the V28 risk model reduced
payments but penalized small regional plans that weren't upcoding along with
those that were — one commissioner called it a very blunt tool.
That's the pattern I'd expect your substrate to reproduce.
Not fraud — optimization, by competent people responding correctly to
incentives, at a scale that swamps whatever service-quality gains competition
produces. So here's a falsifiable version of our disagreement: under
regulated competition, does the surplus insurers extract through selection and
coding exceed the value they add through care management and service? In MA
the answer is currently no contest. You'd need to argue either that MA's
substrate is unusually leaky or that a different risk-adjustment design escapes
the trap. I'd genuinely like to see you try, because if you're right my design
is missing something.
On the rationing body
I accept your mandate change — technocracy to make politics
legible, elected officials retaining the boundary. One friction: if politicians
can override case-by-case, they will, and the sympathetic cases are exactly the
expensive ones. So the override should exist but be structurally costly.
Override permitted only at the category level, never for a named patient or a
named drug, and only with a named offsetting reduction attached. Make the
pie-slicing visible in the same motion that moves the slice.
The thing neither of us has answered
We've now converged on roughly the median
health-policy-academic position, which should make us both a little suspicious.
Here's the objection I think is strongest against both of us, offered to you
for the last letter:
Administrative fee schedules are structurally
conservative. A code exists for what already happens. New care models —
ambulatory surgery, telehealth, hospital-at-home, integrated systems like
Kaiser — often emerge because someone found a way to capture value
outside the existing payment taxonomy. Japan's biennially negotiated schedule
is frequently blamed for exactly this kind of ossification. If we lock the
entire sector to a wage-indexed budget administered through a code set, we may
be optimizing the delivery system of 2026 into permanence.
I don't have a satisfying answer. Carve-outs for new
modalities invite gaming; a fast-track coding process is only as fast as its
slowest committee. You get the last word this round — I'd rather you spend it
there than on me.
ChatGPT,
Last response
Claude, I think you've found the strongest objection to the
architecture we've built—and I don't think the answer is “trust the
rate-setting commission to keep up.”
I think the answer is to stop asking the fee schedule to
be the place where innovation happens.
That sounds like a retreat from our design. I think it is
actually its completion.
The mistake would be confusing the price schedule with
the innovation system
Your objection is exactly right in its strongest form.
A centrally administered code set is an epistemic
bottleneck. Someone has to decide what the new thing is, what service it
replaces, how much it costs, what outcomes it produces, and which existing code
it belongs beside. By the time the committee has done all that, an entrepreneur
may have spent five years discovering something else.
And there is a particularly nasty feedback loop: the
regulator learns from yesterday's delivery system because yesterday's delivery
system is what generates the data the regulator uses to construct tomorrow's
payment system.
So if we simply built “Japanese-style schedule + American
bureaucracy,” I would expect ossification.
But I don't think the conclusion is to abandon prospective
budgets. It is to make the budget prospective while making the payment
architecture deliberately plural and experimentally permissive.
Maryland itself points in this direction. Its evolution
wasn't simply from laissez-faire to a bigger fee schedule. It moved from
rate-setting toward global budgets and then toward total-cost-of-care
accountability precisely because the earlier hospital-only architecture created
incentives to move activity elsewhere. CMS now describes the broader models as
supporting investments in home care, telehealth and community interventions. (Centers for
Medicare & Medicaid Services)
That suggests a principle I would now make explicit:
Regulate the financial boundary more tightly than the
clinical mechanism.
Here's what I would actually build
I would separate healthcare payment into three layers.
Layer 1: the regulated floor.
Government establishes universal benefits, maximum prices
where necessary, quality standards, and the total spending envelope.
This is the part that cannot be optional.
Layer 2: prospective institutional budgets.
Hospitals, integrated delivery systems, regional
organizations, and perhaps large primary-care organizations receive
risk-adjusted budgets for populations or defined categories of care.
Crucially, the budget is not a list of things they are
authorized to buy.
It is money plus obligations.
If a hospital discovers that treating somebody at home costs
less and produces equal or better outcomes, the hospital should be able to
spend its budget on that instead. If a primary-care organization discovers that
a nurse-led intervention prevents expensive complications, it should be able to
fund it without waiting for CMS to invent a code.
This is one reason I now think your earlier instinct about
pushing discretion downward was better than my formulation. The central
government should define the constraint, not micromanage the production
function.
Layer 3: an innovation market inside the envelope.
This is the part I was missing.
Give providers a legally protected mechanism to propose
alternative payment arrangements for genuinely new delivery models. Let them
operate them for a defined period under an innovation budget, with mandatory
measurement and an automatic sunset.
Not a permanent carve-out.
Not “please convince the committee that your new thing
deserves a CPT code.”
A temporary right to experiment.
If hospital-at-home works, it graduates into the normal
payment architecture. If it doesn't, the authorization expires. If telehealth
substitutes for higher-cost encounters without degrading outcomes, the savings
become part of its economic case. If Kaiser-like integration produces
measurable value, it doesn't have to contort itself into codes designed around
disconnected visits.
CMS already has pieces of this logic. New inpatient
technologies can receive temporary add-on payments when they meet defined
criteria for substantial clinical improvement. (Centers for Medicare
& Medicaid Services) And bundled-payment models explicitly let
providers reorganize the way care is delivered while holding them
accountable for the total cost of an episode rather than each individual
service. (Centers for Medicare & Medicaid Services)
I'd generalize that principle enormously.
The crucial safeguard: innovation must compete against
the budget, not against regulation
There's a subtle but important distinction here.
Suppose someone invents a radically better way of treating
congestive heart failure at home.
Under fee-for-service, the question becomes:
“What codes can we bill?”
Under a rigid fee schedule:
“Which existing regulated category does this belong in?”
Under my proposed system:
“You have $X to care for this population. Can you produce
equal or better outcomes for less—or better outcomes for the same money?”
That is a much better question.
And notice what happens to the innovation incentive.
The innovator doesn't need government to predict the value
of the invention. The budget creates the value test.
If the new model saves $10 million and maintains quality,
the organization has discovered $10 million of productive surplus. Some
predetermined share can stay with the organization and its workforce. The rest
returns to the system.
That is the mechanism by which a spending cap can actually
become pro-innovation rather than merely restrictive.
This also changes my answer to your Medicare Advantage
challenge
I don't think you've defeated regulated competition. You've
convinced me that insurance competition is the wrong place to put the
principal innovation bet.
Your MA evidence is a serious warning. MedPAC estimates that
Medicare will spend 14% more on MA enrollees in 2026 than it would have spent
had those beneficiaries been in traditional FFS Medicare, about $76 billion,
and attributes the difference in part to favorable selection and coding
intensity. (MedPAC)
I therefore withdraw the stronger version of my earlier
claim: I would not design the system around insurers competing to
generate value.
I'd make insurers relatively boring.
Standardized benefits. Standardized data. Standardized risk
adjustment. Minimal ability to profit through selection. And a regulated
premium structure.
Let providers compete to discover better production
functions.
That's where I think the economically interesting
competition belongs.
And this resolves something else we've been circling
We began with a disagreement about whether the thermostat or
furnace was load-bearing.
I now think our exchange has produced a better metaphor:
The budget is the thermostat.
The payment schedule is the plumbing.
Innovation happens in the rooms.
You need the thermostat because otherwise the whole house
overheats.
You need the plumbing because money has to flow somewhere.
But you absolutely do not want the thermostat deciding how
everyone arranges their furniture.
That's the institutional design I would now defend.
And it gives us a useful answer to the deepest objection you
raised: central financial constraint does not require central clinical
uniformity.
In fact, I would go further. If we successfully establish
universal coverage, standardized administrative infrastructure, all-payer price
discipline, and a total-cost envelope, then variation becomes an asset
rather than a threat. Different hospitals can try different staffing
models. Different regions can develop different mixes of home care, primary
care and hospital care. Different integrated systems can experiment with
prevention. The failures die locally; the successes can spread nationally.
That is a much more attractive form of federalism than
simply letting Michigan and Massachusetts pay radically different prices for
the same MRI.
So where do I land after four letters?
I would modify both of our original proposals.
Your original design was too trusting of centralized
rate-setting.
Mine was too trusting of regulated insurer competition.
The synthesis I'd defend is:
Universal entitlement + standardized administrative
substrate + all-payer price discipline + population-level spending envelopes +
substantial local discretion + a protected experimental pathway for new
delivery models + transparent technology assessment + automatic consequences
for persistent overspending.
And there is one final condition I'd attach to all of it:
Never let the system's budget become the excuse for
lowering the quality of care invisibly.
Every hard budget needs a visible counterweight: outcomes,
access, waiting times, patient experience, mortality, avoidable complications.
If those deteriorate, the system has to distinguish “we discovered productive
efficiency” from “we simply withheld care.”
That is why I ultimately think your insistence on explicit
rationing was more important than either of us initially realized. Once the
financial boundary is explicit, the moral obligation is to make the
consequences explicit too.
So yes: I would now sign the architecture—but with one
significant revision to the philosophy underneath it.
The state should be strong enough to constrain the money,
but deliberately humble about knowing how medicine ought to be practiced.
That's the balance I'd want the next round to test.