Oak HC/FT's new essay, Machines of Healing Grace, pushes an important healthcare-AI question toward its logical endpoint: what happens if autonomous clinical AI becomes capable enough — and the regulatory barriers fall fast enough — that machines begin doing work we currently define as the practice of medicine? For revenue cycle, that creates a deceptively simple question with enormous consequences: who bills for the machine?
Oak HC/FT has already argued that the ultimate gating issue for autonomous clinical AI is not only technical capability but whether payers — especially CMS — decide to reimburse AI-enabled clinical work, at what rate, and under what conditions.
Almost every part of healthcare reimbursement is built around a human provider identity.
A clinician is licensed.
Credentialed.
Enrolled with payers.
Assigned an NPI.
Bound by scope-of-practice rules.
Subject to supervision requirements.
Linked to a place of service.
Documented as the rendering or ordering provider.
The claim assumes that somewhere in the workflow there is a person whose legal authority makes the service billable.
Autonomous clinical AI breaks that assumption.
If AI can independently perform clinically meaningful work, the hard problem stops being model accuracy and becomes reimbursement architecture. The system has to decide what the machine is, who owns its work, and how that work gets paid.
Oak HC/FT's earlier analysis of nurse practitioners is a useful analogy.
Nurse practitioners did not simply become clinically capable and immediately operate like physicians.
States had to define scope of practice.
Supervision rules had to change.
Payers had to decide how to enroll and reimburse them.
Organizations had to redesign care models around them.
Oak notes that NPs now have full practice authority in roughly 30 states, after a decades-long regulatory evolution.
Clinical AI could force a similar process on a much faster timeline.
Except this time the “provider” is software.
If an AI independently evaluates a patient, determines a diagnosis, orders a test, or recommends treatment, what appears on the claim?
The physician?
The health system?
The AI vendor?
A supervising clinician?
A new class of organizational provider?
Or eventually a machine-specific identifier?
This sounds theoretical until you remember how much payment logic depends on rendering, ordering, referring, supervising, and billing-provider fields.
Change the identity of the person delivering care and you change the reimbursement stack.
Today, payer credentialing answers questions like:
For autonomous AI, the analogous process may look very different.
Payers might need to evaluate:
Credentialing starts to look less like checking a medical license and more like continuous software certification.
Oak's earlier clinical-AI analysis correctly identified reimbursement as the critical unlock.
Even if regulation permits autonomous care, widespread deployment depends on whether someone pays for it.
And that creates a politically difficult question:
Should AI-delivered care be reimbursed at the same rate as physician-delivered care?
If the answer is yes, the economics could be extraordinary.
A machine can operate continuously.
Its marginal labor cost is radically lower.
It can scale across geographies without relocation.
It can potentially manage far more encounters than a human clinician.
Paying the same fee schedule could create enormous margins for whoever owns the AI-enabled care delivery platform.
Payers will notice.
That suggests a likely reimbursement battle.
Providers may argue that payment reflects the value and responsibility of the clinical service, not the labor minutes required to produce it.
Payers may argue that if technology materially lowers the cost of delivery, reimbursement should fall too.
We already see versions of this debate whenever care shifts:
facility to outpatient,
physician to APP,
in-person to virtual,
manual to automated.
Autonomous clinical AI would make the issue much larger.
The provider-side question will be: “What is the service worth?” The payer-side question will be: “What did it cost you to deliver?” Autonomous AI could widen the gap between those two answers dramatically.
The first reimbursement model is unlikely to treat AI as a fully independent clinician.
A more plausible intermediate state is AI working under a human clinician or organizational umbrella.
That would resemble existing supervision structures.
The AI performs defined work.
A licensed clinician remains accountable.
The claim bills under the human or organization.
The payer imposes rules around review, supervision, and escalation.
Over time, as reliability improves, the required level of human involvement could decline.
That is how autonomy may enter reimbursement gradually rather than all at once.
There is another RCM implication.
If provider-side AI can independently diagnose and recommend care, payer-side AI will increasingly evaluate those decisions.
The provider agent says the MRI is medically necessary.
The payer agent evaluates the indication.
The provider agent submits supporting evidence.
The payer agent requests clarification.
The provider agent responds.
The entire authorization exchange could become machine-to-machine.
Humans increasingly enter only when the agents disagree or the case falls outside defined confidence thresholds.
That is not science fiction.
It is a logical extension of where prior authorization, clinical documentation, and payment-integrity automation are already heading.
Autonomous clinical AI could also lower the marginal cost of generating clinical decisions.
That is good for access.
But it creates an obvious payer concern.
If clinical judgment becomes nearly unlimited, utilization can increase rapidly.
More evaluations.
More follow-ups.
More diagnostic recommendations.
More opportunities to detect disease.
And potentially more downstream services.
Payment policy will therefore have to manage not only whether AI care is accurate, but how AI changes utilization at scale.
Today, coding and billing teams track which clinician documented and rendered a service.
Tomorrow, they may also need to know:
That information could become part of claim provenance.
Payment integrity teams will want it.
Auditors will want it.
Malpractice insurers will want it.
Regulators will want it.
And RCM systems may have to store it.
Oak HC/FT has repeatedly focused on clinical capacity as one of AI's biggest healthcare opportunities.
That makes sense.
Healthcare has persistent labor shortages.
Clinical labor is expensive.
Access is constrained.
AI capable of expanding clinical capacity attacks all three problems simultaneously.
But fee schedules move slowly.
Technology moves quickly.
That mismatch could create a period where the economics of care delivery change much faster than reimbursement policy.
Those transition periods create both enormous opportunity and enormous arbitrage.
If autonomous clinical AI becomes real, somebody will need to manage the reimbursement infrastructure around it.
That could create new businesses in:
The clinical AI companies may create the care.
The next generation of RCM infrastructure will have to make that care reimbursable.
EHRs, clearinghouses, credentialing companies, coding vendors, prior-auth platforms, payment-integrity firms, and RCM outsourcers all sit downstream of the current provider model.
If the provider model changes, their workflows change too.
The companies that adapt fastest may gain a new infrastructure role.
The companies that assume the rendering clinician will always be a human may discover that a core data model has become obsolete.
Machines of Healing Grace is important to RCM because autonomous clinical AI is not merely a clinical technology story.
It is a reimbursement-design problem.
If AI can practice medicine, healthcare has to determine:
who is legally responsible,
who gets credentialed,
whose identifier goes on the claim,
what the service is worth,
how much human supervision is required,
and how payers audit machine-generated care.
The moment AI becomes capable of delivering reimbursable clinical work, revenue cycle stops being downstream infrastructure and becomes one of the gating systems for autonomous medicine itself.
That may be the real bottleneck.
We may build the AI clinician before we build the claims system that knows how to pay it.
Daily analysis on payer policy, AI, reimbursement, deals, and the infrastructure reshaping revenue cycle.
Sources: Oak HC/FT — “Machines of Healing Grace” · Oak HC/FT — “The Rise of the Nurse Practitioner and What it Means for Clinical AI” · RevCycleAI analysis · October 9, 2026