AI is coming for everything broken in healthcare.

Not just the inefficiencies and the middlemen — although I’d love those to be first. It’s coming for perverse incentives, and outdated payment structures, and bureaucracy and paperwork and administrator meetings. *Hallelujah. *

AI doesn't break these systems outright, and it doesn’t rip a hole in your couch. But it exposes all the weakened seams. It might cut a few of them, but I think it mostly applies pressure exactly where the logic was already failing, and unravels the stitching that we’ve been pretending wasn’t ever the problem.

And while fee-for-service isn’t first on the chopping block either, I think it’s one worth discussing because it's so fundamentally tied to so much of American healthcare. And what will happen when AI starts doing a healthy chunk of the stuff that we usually pay doctors to do?

Why Fee-for-Service Works Everywhere Else — and Breaks in Medicine

Us policy’n’payment nerds claimed the term "fee-for-service” for healthcare, but let’s call out that it’s a pretty good descriptor for a large portion of how you and I spend our money. It works fine for your mechanic, your Netflix, and your local pizza place.

You have a problem or a need → someone fixes it → you pay for the fix. Clear deliverable. Observable quality. Market discipline. If the repair doesn't hold or the pizza order’s wrong, you stop being a customer, you write a review, you request a refund. Competition keeps prices in check. Also important: I pay, I decide, you pay, you decide.

Medicine sticks out like a sore thumb. In healthcare:

  • Information asymmetry is extreme. The person recommending the service is the one who profits from it — and you often can't tell if the recommendation was appropriate. Not before, not during, sometimes not even after.
  • Third-party payment severs the feedback loop. The person receiving care isn't the person paying. When your mechanic quotes $800 — oof — you feel it. When your doctor orders an MRI, your insurer pays $1,200 and you pay $100. Normal price signals never reach the consumer.
  • Quality is often unobservable. Did the treatment work, or would you have gotten better anyway? Was the test necessary, or defensive? Was the surgery indicated, or preference-sensitive? Even sophisticated patients often can't tell. Even sophisticated, experienced doctors can’t tell if it was “necessary” — often until the test is completed.
  • Outcomes are probabilistic and delayed. A plumber either fixes the leak or doesn't — you know within hours. Medical interventions have percentage chances of working, and the outcome may not be apparent for months or years. Other outcomes aren’t fully under anyone’s control but god and the universe, and we’re extremely price insensitive for many of them in healthcare.
  • Demand is inelastic and urgent. Who’s shopping around when their aneurysm ruptures? The usual market discipline — walk away, find another vendor — doesn't apply when the alternative is death or disability.
  • Chronic disease has no "completion." FFS assumes a discrete transaction: problem → fix → done. And this made sense 100 years ago when most of medicine was acute problems with a beginning, a treatment, and an end. But diabetes, hypertension, heart failure, depression — these aren't fixed. They're managed. Indefinitely. FFS has no clean way to pay for "keep this person stable and out of the hospital” besides “keep paying for intermittent services in a continuous world.”
  • The most valuable work is invisible and provides shared value to all of us. Prevention, coordination, patient education, medication reconciliation, goals-of-care conversations — these produce no "thing" to bill for, yet are arguably useful to both individuals and everyone else in society. Fewer strokes means less disability and more bandwidth for the healthcare system to manage other stuff. FFS underpays the work that prevents the expensive downstream disaster.

Fee-for-service assumes a world of clear deliverables and informed buyers. But I can’t go on Consumer Reports or RTINGS.com and figure out which doctor or hospital or blood pressure pill is best, like I can with a new flatscreen TV.

You know where this lands: FFS in healthcare creates incentives to do more, regardless of whether more is better. (NB: Sometimes more is better.) But some of the most valuable work — the conversation that avoids a futile ICU admission, the coordination that prevents the readmission — produces no "thing" to bill for. Sure, we’ve patched it up with a bunch of drywall mud many times over the years, but like, how often can the landlord do that before the wall itself can’t hold up a picture frame?

AI didn't create this mismatch. But I think it’s already starting to make it undeniable, and we’re still just skimming the surface.

Where Fee-for-Service Still Works

Before you start raging in the comments, let me stop you. Fee-for-service isn't the villain, but I do think its monopoly is and has been for many decades.

Fee-for-service is actually appropriate for certain kinds of medicine — discrete, well-defined services where the "thing" is clear, bounded, and auditable. Set a fracture. Drain an abscess. Do a colonoscopy. Read a CT. Suture the laceration. In those lanes, FFS can be clean, legible, and even fair. (Ironically a lot of that is common in my field of emergency medicine, although I’m a value-based baddie.)

The problem is when FFS stops being a payment tool for appropriate contexts and becomes the dominant operating system for all of healthcare — especially chronic disease, longitudinal complexity, and everything that happens "between visits."

That's where the logic rots. I think everyone who works in healthcare can feel it, but I want to call it out.

Enter AI: The Revealer

If you come in for a sore throat, I might collect a fee from you (or more likely, your insurer) in exchange for a bunch of services:

  • Take a history
  • Examine you
  • Use my medical expertise to manage/treat you
  • Accept some liability for your care as a licensed professional
  • Write a note about you documenting it all

What happens when more and more of the above tasks are done by software? Right now we’re seeing this widely adopted in medicine for note-writing, which is one of the bundle-of-things that fee-for-service has traditionally paid for, but not others. However I promise you that while I think the others are not yet ready for prime time in January 2026, they are on their way.

“Now Graham,” you say, “the AI scribe might be helping write the note for the doctor, but the doctor’s reviewing it and signing it, and the AI scribe is just a tool.” And you would be right — notes need to exist, doctors should use tools, and notes need to be accurate and documented (I’d argue more to help the patient’s future care than anything else — but that’s a separate topic).

But my point is that we're drifting into a world where we keep paying as if the same human labor happened — even when the work is increasingly machine-generated — and we haven't decided what the payment model should become instead (and, more importantly, what is the doctor doing instead).

So I’ll get back to my titular point: if the machine’s providing the service, what’s the doctor’s fee for?

Digital Deflation is Already Here

Fee-for-service makes sense in acute, episodic, fixed-beginning-and-end care. But it also acknowledges that work takes time, and allows us to pay different amounts for different things, as opposed to the legal profession, for example, where everything is just billed by the minute. Open heart surgery has a higher per-minute value than a hangnail.

But when the note becomes cheap to produce, doesn’t AI create digital deflation? If the marginal cost of producing "units" of clinical work product collapses — notes, summaries, coding suggestions, triage flags, draft plans over time — doesn’t that make AI capable of “printing cheaper clinical minutes” just like the Federal reserve can print money?

I think CMS already knows this is a problem. In the CY 2026 Physician Fee Schedule final rule, CMS explicitly notes that time assumptions used to value codes may be overinflated, and finalizes an "efficiency adjustment" to work RVUs for many non–time-based services. That's the Federal government's payment system admitting out loud: if practice gets faster, the unit price can't stay frozen forever.

And I also think CMS is trying to move the Titanic-sized Medicare system away from the iceberg of this question.

Is CMS Stress-Testing the Future?

This isn't just a thought experiment. CMS's Innovation Center has launched models (that I can barely keep up with) that are real-world disruptions into exactly this thesis: stop paying for discrete units of activity and start paying for outcomes, trajectories, and accountability — especially as technology makes the activity cheaper to generate.

  • ACCESStests outcome-aligned payment in Original Medicare (importantly, fee-for-service-only) for technology-supported care in hypertension, diabetes, chronic musculoskeletal pain, and depression. It's explicitly designed to pay for outcomes over activities.

That's CMS saying out loud what fee-for-service struggles to say: a lot of modern care is continuous, hybrid (in-person + asynchronous + device + coaching), and not naturally divisible into billable discrete units.

  • LEAD (Long-term Enhanced ACO Design) is accountable care architecture — population-based payments, prevention, coordination, flexibility "between visits."
  • ASM (Ambulatory Specialty Model) applies the same logic to specialists: mandatory participation in selected regions for certain outpatient specialists treating heart failure or low back pain, starting January 2027, aimed at upstream management and reducing avoidable hospitalizations.

So yes — if AI keeps compressing the cost of producing clinical "stuff” (and news flash: it’s going to), the physics point toward models like ACCESS, ACO structures like LEAD, and specialty accountability experiments like ASM.

CPAs Faced This Decades Ago

Medicine isn't the first profession to confront this. (I’m intentionally avoiding “Airlines” and “pilots” and “auto-pilot” analogies here because it’s been so over-done.)

Tax preparation software automated calculation, reconciliation, and basic returns. TurboTax commoditized the low end. But CPAs didn't disappear — many just moved up the chain.

What changed was what they're paid for. Not arithmetic. Not form completion. CPAs now bill for tax strategy, compliance interpretation, audit defense, and judgment calls on ambiguous situations. (Are they happier? Let me know in the comments.) The production of documents became nearly free and completed by individuals entering their details into a webpage. Even some of the simpler tax logic questions got automated in TurboTax, but the nuance and complicated situations didn’t go away.

The pattern: when automation handles production, professionals get paid for judgment, relationships, and accountability. Hmm. *Y’know, doctors do a lot of those things too. *

Documentation is the calculator arithmetic. Clinical reasoning is the advisory work. If medicine follows accounting, the "service" shifts from producing notes to interpreting situations, managing uncertainty, and bearing responsibility when things go sideways.

Value-Based Care Is, I Think, Better, But Ain't Salvation Either

The cleanest conceptual answer is pay less for production and more for outcomes, stewardship, and accountability. That can look like capitation, bundles, shared savings, or condition-specific hybrids. (I can feel you rolling your eyes, thinking “Of course the Permanente doc is going to say this!”)

But here's the part people skip: AI doesn't just fit value-based care. It creates new failure modes, too.

If you pay for outcomes, AI will optimize to the metric. Sometimes that's great. Sometimes it’s a mess that creates its own challenges.

But I do think outcomes-based models naturally align incentives around efficiency and quality — AI is easier to adopt when you're not paid per unit of human labor. And outcomes importantly also align with patients, which is, as a reminder, the whole point of the healthcare thing.

Necessity is the mother of invention, and I’ve seen tremendous invention and innovation when you can ask yourself, “I wonder if there’s a better — but just as safe — way to do this for my patient? And perhaps could it also be cheaper, too?"

Hopefully you agree that AI likely accelerates the move away from fee-for-service dominance, because it starts to question the idea of paying the human to do the thing. CMMI's portfolio is clearly leaning that direction. But it’s also clear to me that AI doesn’t just snap its cloud-based fingers and fix everything. AI just changes where we need new resources:

  • governance stakes in value-based care
  • who sets the tradeoffs
  • who audits the models, and
  • who is accountable when the algorithm's "win" doesn't feel like a win to the patient.

The Questions That Will Start Fights

Had to throw in a little more spice here, and I don't have these answers:

  • If AI provides the documentation, what is the physician's fee actually for — production, judgment, or being the liability wrapper around the output? If us doctors are just an accountability and liability sponge for everyone else in the system, then who’s gonna sign up for that job in 10 years?
  • And if we move toward outcomes-based care, what guardrails stop AI from turning "value" into a spreadsheet and dashboard victory and a human loss? There's already hundreds of "quality metrics." How many more can we take?

  • Finally: Who gets the AI dividend? If AI cuts labor time and reduces friction, that makes surplus. Who gets that?

  • Patients (lower costs, better access)

  • Payers and Employers (lower spend)
  • Providers (same payment, lower internal cost)
  • Vendors (software rent, plus eventual reimbursement leverage?)

The Bottom Line

Fee-for-service was built when physician time was the scarce input. AI makes the artifacts of that time — notes, summaries, triage flags — cheap to produce.

CPAs survived this. They shifted from arithmetic to advisory work: judgment, interpretation, accountability. Medicine has the same opportunity — but only if we stop paying for documentation and start paying for what patients actually need.

The question isn't whether payment will change. It will. The question is whether physicians get paid for judgment and trust — or just end up holding the liability while the savings go everywhere else.

Thanks to my incredible colleague and Chief Innovation Officer Kenneth Trauner MD for us spitballing this idea 6 months ago and Kayleigh-Ann Clegg's interview questions that motivated me to get this idea out of my brain and onto paper!