Free email sprint · 3 days
Build your FDA pathway argument — and find the one feature that forces a harder pathway — in 3 prompts, over 3 days.
“Generative” doesn't force De Novo. What forces a harder, slower pathway is letting the model hold the clinical decision — and the teams that clear fastest bound the AI around a deterministic core and ride a predicate. This sprint builds that argument for your device, then finds the feature most likely to break it.
- Three short emails, one prompt each — run in your own AI platform, against your own device.
- You leave with a drafted pathway argument and your single biggest pathway risk.
- Written by Ross Prior (Saolyn), from The State of AI in Medical Devices 2026 with Kelly Coverdale (Cover Biomedical).
Opening shortly
The sprint isn't open for signup yet. In the meantime, the report it's drawn from is on the reports page.
See the reportWhat the three days cover
- 1
Map where your model actually decides
The pathway turns almost entirely on whether the AI holds final clinical-decision authority, or whether a deterministic rule or a human sits between the model and the patient. Day one maps your device function by function and finds where, if anywhere, it decides unsupervised.
- 2
Build the pathway and the predicate case
A named pathway, and the part teams hand-wave: the legally-marketed device you're substantially equivalent to, on what grounds, and the evidence a reviewer will expect. “Probably a 510(k)” is not a strategy.
- 3
Find the pathway-breaker, and bound it
The single feature most likely to force a harder, slower route — and a concrete way to wrap it around a deterministic or human-supervised step, with the trade-off named rather than discovered in review.