Software now writes the medicine's paperwork
AI already drafts prior-authorization letters, coverage determinations, visit notes, medical-necessity letters, and care plans. It's fast, it's cheap, and a lot of the time it's good. Healthcare AI companies are shipping these tools into payers, health systems, and clinics right now.
But a draft is not a decision. Software can generate a conclusion. It cannot be accountable for one — and in medicine, someone always has to be.
An algorithm can assist. It cannot decide.
This isn't a values argument anymore — it's written into the rules.
CMS — Medicare Advantage Final Rule (2024)
A Medicare Advantage plan may not base a coverage or medical-necessity denial on an algorithm or AI alone. The decision must rest on the individual patient's circumstances — their history, the treating physician's recommendation, the clinical notes — and cutting off care requires a qualified human to re-assess that specific patient.
The states — a 2026 prior-authorization wave
A growing list of states now require that a licensed, specialty-matched clinician — not a model — review and own a medical-necessity denial, on the individual record. The direction across CMS and the states is uniform: AI may draft; a named licensed human must decide and sign.
Read that again as an AI company: every consequential output your model produces now needs a licensed, specialty-matched human to stand behind it — in every state you sell into.
The AI companies don't have the physicians
To sell into a regulated world, an AI vendor needs licensed, specialty-matched physicians who will review its outputs and sign. Building that in-house — a national, multi-specialty, credentialed panel with malpractice cover and an audit trail — is slow, expensive, and outside what an engineering team knows how to do.
The model is the commodity. The accountable human signature is the product — and it's the part nobody has built at scale.
The physician layer, delivered as an API
An AI company submits an output. ClinicalSwipe routes it to an NPI-verified physician in the matching specialty. The physician reviews it against the source, then approves, edits, or rejects it — and signs. The output leaves carrying a signed record and a 7-year audit trail: proof that a qualified human owned the decision, exactly as the law requires.
One integration for the AI company. For you, a stream of reviews in your specialty. In the middle, the thing the law now demands.
You review the work. Your judgment is what's for sale.
From your phone, on your schedule, you swipe through AI-drafted clinical work in your specialty — approve, edit, or reject. Each decision is timestamped and attributed to you. You are the accountable, licensed human the law requires; your signature is what makes the AI usable at all.
To be clear about the claim: your value isn't making the AI "more accurate." It's accountability — being the named, licensed clinician who reviewed the individual case and stands behind it. That's what CMS and the states are actually asking for.
Early, and built in the open
This is a founding cohort, not a finished company. It's physician-built and hands-on — the review experience is live, the first reviewers are coming on now, and the commercial side (the AI companies who pay for reviews) is the work ahead.
That's the opportunity: you'd be shaping how physician review actually works in the AI era — the specialties, the standards, the workflow — from the ground floor, not joining something already set in concrete.