clinicians.build · interactive · august 18, 2026

The Counted

Ask a health system how many AI agents it has deployed and you’ll hear 50. Run the audit and it’s 76. Here is the entire universe of clinical AI that the federal government actually counts — 125 clearances, eleven years — so you can see how small the counted part is.

Primary source: The 229 Podcast (This Week Health), “50 Agents on Paper. An Audit Finds 76.” — Newsday with 229Project, 17 Aug 2026
Grounding data: MIMI Labs · FDA 510(k) Premarket Notification database joined to the FDA product-code classification file, decisions 1 Jan 2015 – 31 Jul 2026

Sarah Richardson’s line on this week’s Newsday: ask a health system how many agents it’s running, then run the audit and add fifty percent. Fifty on the slide, seventy-six in the building. Bill Russell’s sharper follow-on from the same segment — most of that AI arrived through Workday, ServiceNow and the EHR, not through anything resembling procurement.

So it’s worth knowing what the counted universe looks like, because there is one. FDA runs a product-code taxonomy, and a handful of those codes explicitly name computer-aided detection, machine learning, or artificial intelligence. Every 510(k) cleared under one of them is a clinical AI product that a federal agency has a file on.

Every 510(k) cleared under an AI/CAD/ML product code, Jan 2015 – Jul 2026
Clearances
125
Distinct firms
69
Product codes in play
15
Radiology’s share
89.6%
One hundred twenty-five clearances is the whole federal record of AI-coded devices for the entire United States over eleven and a half years. The 229 audit found 26 uncounted agents inside a single health system. That one system’s surplus is 21% of the national eleven-year total — and almost none of those 26 are devices at all, so none of them would ever appear here.

All 125, plotted by product code and clearance date

Each circle is one clearance. Vertical position is the FDA product code it was cleared under; horizontal position is the decision date. Hover any dot for the device, the firm, and the K number.

≥1
Radiology Cardiovascular Other — rows below the floor grey out
Shown
125
Firms
69
Codes surviving floor
15
Top firm’s share
13.6%
Drag the sample‑size floor. Switch Clearances to Distinct product families.

Two things the raw count hides

It isn’t 125 products. Flip the counter to distinct product families — the same firm re-clearing the same trade name is one product with a version history, not a new capability. Aidoc alone accounts for 17 of the 125 clearances, nearly all of them BriefCase. The headline number is a filing count wearing an innovation costume.

It isn’t medicine. Push the floor past a handful and everything outside two specialties disappears. Pathology has one clearance in eleven years. Neurology has one, and it is a hair-transplant robot. If your slide says clinical AI is diffusing across the house of medicine, the federal record says it is diffusing across radiology and a corner of cardiology.

What a real inventory would have to hold

The chart above spans eleven and a half years, the entire country, and every firm that has ever cleared an AI-coded device. Below, the same 125 against what one health system found in one afternoon.

The 50-vs-76 figures are one executive’s recounted audit on a podcast, not a published study — treat them as an anecdote with a good shape, not a rate. The 125 is exact and reproducible from the FDA file. Comparing them is a scale illustration, not an apples-to-apples measurement: one counts regulated devices nationally, the other counts software agents at one organisation.

Your tool gets counted whether or not anyone tells you. Ship the inventory line before the audit asks for it.