The Bill and the Answer

Two pieces of paper on one kitchen table, and the quietest question in American politics: what does care actually cost to make? Not medical advice, not financial advice, not a political piece — a walk around a tension worth thinking about.

Picture a father at a kitchen table — a composite, stitched from the public record rather than any real family, but you know him. His daughter spent four hours in an emergency room last month with a bad asthma flare. She’s fine. The bill arrived today: four figures, after insurance, itemized in a language no one at the table speaks.

That evening, worried about her discharge instructions, he types a question into the AI on his phone. In eight seconds he gets a patient, expert-level explanation of her medications, warning signs, and follow-up questions. It costs him nothing. Five years ago, that conversation was a copay and a three-week wait, if it happened at all.

Two pieces of paper on one table. The answer — the expertise — has become nearly free. The bill has not. The distance between them is becoming one of the clearest tests this century will run: when intelligence gets cheap, does the cheapness reach people, or does it get absorbed on the way?

Let’s walk around it slowly, jerseys left at the door.

Two Curves, One Country

Start with what’s measured.

American health spending reached $5.3 trillion in 2024 — 18.0% of GDP, up from 17.6% the year before. In the two decades after 2000, as televisions fell 97% in price, hospital services rose about 220% — the famous “chart of the century.” Downstream: Americans owe at least $220 billion in medical debt; about 20 million people carry some, and roughly 3 million owe more than $10,000. That’s the father at the table, multiplied by twenty million.

Now the other curve. The FDA has authorized over a thousand AI-enabled medical devices — a taxonomy in npj Digital Medicine counted 1,016 authorizations, most in imaging. The first autonomous AI diagnostic — detecting diabetic retinopathy in a primary-care office, no specialist present — cleared its pivotal trial with 87.2% sensitivity and 90.7% specificity in 2018. And the raw input beneath it all has been deflating at a rate without precedent: LLM inference for constant capability fell roughly 1,000-fold in three years.

Hold both curves at once; both are real. The knowledge component of medicine — the thing that took a decade of training to carry in one human head — is riding the fastest cost collapse ever recorded. And the bill is riding the same stubborn escalator as ever.

Here’s why this is a policy question: for half a century, America’s healthcare war has been fought almost entirely over who pays — single-payer or market. Both armies quietly share an assumption so deep nobody says it aloud: that the underlying cost of care is roughly fixed, a boulder to be shoved onto someone else’s ledger. AI is the first force in this argument’s history that attacks the boulder itself — but only part of it, and only if the savings actually arrive.

Both sides of the old war have something true to say about that. Both deserve their strongest telling.

The Strongest Case from the Left: Some Doors Shouldn’t Have a Coin Slot

The left’s case begins where economics itself concedes ground. In 1963, Kenneth Arrow — later a Nobel laureate, and no radical — published “Uncertainty and the Welfare Economics of Medical Care,” the founding document of health economics. Its argument still stands: medicine violates the assumptions that make markets work. You don’t know what you’re buying (the doctor does). You can’t comparison-shop mid-heart-attack. You buy rarely, under duress, with your life as collateral. From that root, three branches:

The moral floor. The United States is the only high-income country without universal health coverage. One camp of researchers finds medical problems contribute to about two-thirds of personal bankruptcies; a rival team of economists, using stricter causal methods on a narrower question (bankruptcies caused by a hospitalization), puts that figure closer to 4%. The honest reading: the true number sits in a wide, contested range — and no other rich country needs the debate at all, because no other rich country lets sickness reach the bankruptcy court. The left’s deepest claim isn’t statistical: some doors should never have a coin slot. A child’s insulin. A bed when it’s freezing.

The paperwork tax. Administration consumed 34.2% of U.S. health spending in 2017 — roughly $812 billion — versus 17% in Canada, by the most-cited estimate (from Himmelstein and Woolhandler, who favor single-payer; critics note their accounting runs generous). A more conservative JAMA tally put administrative complexity at $265.6 billion a year in pure waste — the largest category in a study counting $760–935 billion of total annual waste. Even the low estimate is the size of a small European economy — a cost, the left argues, created by the who-pays architecture itself.

Negotiating power works. After decades of statutory prohibition, Medicare began negotiating drug prices: the first ten negotiated prices took effect January 1, 2026, with fifteen more — including the GLP-1 blockbusters — set for 2027. The mechanism every other rich country uses is now running here.

That’s the left’s case, honestly told: markets fail in medicine in documented ways, the failure lands on the sick, and pooled public power demonstrably lowers prices. Hold it while we turn the table.

The Strongest Case from the Right: The Cage Around Supply

The right’s case begins with an observation the left rarely engages: the parts of medicine that behave like markets, deflate like markets. LASIK — elective, self-pay, price-advertised — saw the real price of the conventional procedure fall roughly a quarter between 1999 and 2011. Honesty requires the caveats: that’s a modest decline, not a TV-style collapse; many patients paid more for upgraded versions; and an eye procedure you can schedule and shop is nothing like a stroke. The example proves less than its fans claim — and more than its critics allow.

The deeper argument isn’t about shopping. It’s about the cage built around the supply of care — evidence with genuinely bipartisan fingerprints:

Certificate-of-need laws. In 35 states and D.C., a provider needs government permission — which incumbent competitors can formally oppose — to add beds, machines, or facilities. Four decades of research — much of it gathered by the market-oriented Mercatus Center — associates these laws with higher costs, reduced access, and no quality gain; a systematic review concurs.

The residency cap. In 1997, Congress froze Medicare-funded physician training slots at 1996 levels — anticipating a physician surplus. The freeze held for nearly a quarter century (Congress added 1,200 slots in 2021–2023), while the AAMC now projects a shortage of up to 86,000 physicians by 2036. America caps the production of doctors by statute, then wonders why doctors are expensive.

Scope-of-practice walls. Only about 30 states plus D.C. let nurse practitioners practice to the full extent of their training; the rest hold care capacity hostage to turf. Similar walls surround pharmacists and physician assistants.

And innovation runs on returns. The pharmaceutical industry spent about $83 billion on R&D in 2019 — roughly a quarter of revenues — and CBO is blunt that expected future revenue is a primary driver of how many new drugs get developed. Squeeze prices carelessly and you may squeeze the pipeline that makes future care cheap. The right asks the left to say what it rarely says: somebody’s margin funds the miracle.

That’s the right’s case, honestly told: much of the sticky curve isn’t a law of nature — it’s legislated, a lattice of permissions protecting incumbents — and the highest-leverage move is to unlock supply. Serious too.

Notice: these two cases barely touch. One is about the table; one is about the kitchen. They could both be right.

The ReThink: Attack the Cost, Guarantee the Floor

This series measures every reframe against one test: does it (a) move toward abundance, (b) put the most vulnerable at the center, and (c) keep the both-and — the market engine and the dignity floor? Here is a direction to debate — a napkin slid across the table, not a bill dropped on it.

Proposal one: liberalize the supply of care — aggressively, and symmetrically. Let validated AI do what it’s demonstrably safe doing, at the evidence bar the first autonomous diagnostic cleared — pivotal trials and cleared indications, not chatbot vibes. Let nurse practitioners, pharmacists, and physician assistants work to the top of their training everywhere, not in some states. Repeal certificate-of-need. Uncap residencies. And the trans-partisan honesty check: both camps protect supply-side villains — the right defends incumbent hospital systems and guild economics; the left defends credential walls and staffing rigidities its aligned unions ask for. A supply deal is only credible if each side brings its own sacred cow to the table. This is the (a) lever — it makes care cheaper to produce, which no amount of payment reshuffling does.

Proposal two: guarantee the floor — catastrophic and essential care, universally. If supply liberalization is the engine, the floor is the covenant: no American bankrupted by sickness, full stop — a universal backstop for catastrophic costs and a defined basket of essentials, financed publicly, with the market running vigorously above it. This is (b) and (c) in one move: the vulnerable aren’t served by trickle-down from someday-cheaper care, but by a floor that exists before the curves finish their work. The right gets a real market where markets work. The left gets the coin slot pried off the doors that should never have had one. Neither gets everything. That’s what a deal is.

Proposal three: watch the capture. The quiet failure mode that should worry both camps more than each other: the savings arrive and nobody sees them. If AI cuts the cost of reading a scan by 90% and the bill doesn’t move, the abundance happened — on someone’s margin sheet. So treat the two-curves gap as a capture measurement: track the spread between what AI-era care costs to produce and what it’s billed at. Sunlight on the spread is a policy all by itself, and it’s ideologically homeless — which is a compliment.

And name what stays scarce, without flinching. Deflation reaches the diagnosis. It does not reach the hand on the shoulder. Nursing, bedside presence, the vigil at 3 a.m. — care labor is the classic case of Baumol’s cost disease: its product is human time, so its relative price climbs as everything else cheapens. The human parts of care get relatively more expensive forever, and a decent society plans to pay for them — perhaps precisely out of the expertise savings, if we don’t let the savings vanish.

Honest Tradeoffs, Both Directions

A proposal that hides its costs is an advertisement. These are real, and open:

Liberalized supply means errors with new signatures. AI diagnostics fail differently than tired residents do — confidently, at scale, in patterns their training data didn’t cover. The evidence base is real but young and heavily concentrated in imaging; a cleared device is not a cleared system of care; and “demonstrably safe” always needs a where, a whom, and a for-what. Who is liable when an autonomous tool misses — the developer, the clinic, the supervising human who wasn’t required? Nobody has settled it.

The floor’s edges are where the knives come out. What counts as “essential”? Catastrophic-only coverage that ignores the recurring cost quietly bankrupting a diabetic isn’t a floor; it’s a poster of one. Draw the basket too thin and you’ve abandoned people politely; too thick and you’ve rebuilt the comprehensive system you claimed to be simplifying.

Price pressure can eat the pipeline. The CBO’s finding that expected revenue drives drug development doesn’t vanish because it’s inconvenient. If price negotiation and AI-cheapened discovery both land, the net effect on innovation is genuinely uncertain. Anyone who claims to know is selling something.

And the transition has a body count of trust. Millions of livelihoods are, in some sense, the sticky curve. Deflate it clumsily and you don’t get abundance — you get a political immune response that entrenches every wall this essay names.

What to Carry Out the Door

If you shape opinion or policy, three portable things:

One: add a second axis to every healthcare argument you meet. The reflex axis is who-pays. The neglected axis is what-it-costs-to-produce. Proposals that touch only the first are boulder-shoving; the interesting ones from either party shrink the boulder.

Two: judge the next five years by the spread. Expertise costs are collapsing on a public curve. Watch whether any of it reaches a bill — a premium, an ER invoice, a scan price. If by 2030 the answer is still “none,” the problem was never the technology, and the capture question should move to the center of politics.

Three: don’t let the abundance story skip the bedside. The measure of a healthcare system was never the brilliance of its diagnoses. It’s whether the bill stops being a second illness — and whether, when the answer finally gets cheap, someone is still paid, and honored, to hold the hand of the person receiving it.

The fifty-year war asked: who pays? The next one will ask: where did the savings go? Better to start asking now, together, while the answer is still being decided.

Honest Fine Print

This is not medical advice. Nothing here evaluates any treatment, tool, or provider; AI capability claims are limited to cited, peer-reviewed or FDA-cleared evidence. Health decisions belong with licensed clinicians who know you.

This is not financial advice. Nothing here is guidance about insurance, debt, or money; those decisions belong with qualified professionals.

This is not a political verdict. Serious people hold each case above, and both camps include people who lose sleep over the same sick kid. If you finish unable to tell which way this essay votes, it worked.

And this is a proposal, not a decree. Every number is linked to its source; several (administrative costs, medical bankruptcy) are honestly contested, and the ranges are shown. The reframe is offered to be argued with by people who know things we don’t. That’s what the table is for.

Part of the ReThink Policy series. Facts verified against sources current as of July 2026; programs and evidence will evolve.

— The ReThink · firstfruits 🌱 · truth first, hope on top