Fire your $15k coach. Don’t fire the doctor.
The DIY guides going around are right about almost everything: connect your ring, your food log, and your labs to Claude, and skip the concierge markup. Do it. We use the same connectors.
They’re wrong about one thing — and it’s the one that matters. The moment your AI interprets a lab, that’s a clinical act with no doctor and no privacy. Here’s the same build, with the one layer it skips.
Where the DIY stack is dead right
This isn’t a takedown. Most of the build is exactly what we’d tell you to do.
The clinic markup is real.
Paying $15k a year for a concierge to read data you already own is a markup on convenience, not medicine. Connecting your own ring, food log, and labs to an AI is genuinely the right instinct.
How the connect-your-data part works →Your data, one agent.
One AI reading your sleep, HRV, macros, and bloodwork together beats six apps that never talk. The MCP-connector approach the guides teach is sound, and it’s the same plumbing we use.
See it as a dashboard →Awareness beats a binder.
A weekly “here’s the one thing to change” is worth more than a 40-page panel nobody reads. On that, the DIY stack and a good physician completely agree.
Score your own setup →“Here’s what your labs mean and what to ask your doctor” is not a wellness tip. It’s a clinical interpretation.
When a model reads your panel and tells you what’s wrong, a clinical act just happened — with no licensed physician, no NPI attached, and nobody accountable if it’s wrong. And to do it, you handed your full bloodwork to a general AI that never signed an agreement to protect it.
You may be the only one in that conversation who thinks it’s private — and the only one who thinks a doctor was involved. Who owns the conversation? →
The same build, one layer added
Keep every connector the DIY guide gives you. We only change what happens the moment it turns clinical.
Claude reads your sleep, HRV, and recovery. Great — keep this exactly as the guides describe.
No change needed. This is wellness data; on-device or in your own chat is fine. The harness doesn’t touch what doesn’t need touching.
See the on-device layer →Claude reads your macros and trends. Also fine — nutrition tracking is not a clinical act.
Still fine. The line hasn’t been crossed yet. Convenience layer, no doctor required.
Why wellness data is safe to keep →You paste a lab PDF and Claude flags what’s out of range and “what to ask your doctor.” This is where the guide goes quiet: a model just interpreted a clinical result, with no licensed physician and no NPI on it — and your full panel just went to a cloud AI with no BAA.
The governed version (not live yet): a licensed physician reviews the interpretation before you read it, NPI-bound and auditable, with the PHI handled where it is safe. Today you can check any AI summary against your original at the Fidelity Gate.
Check an AI summary today →Claude summarizes the week and names one change. Genuinely useful — automate it.
Automate it too. But when the review touches a medication, a symptom, or a flagged marker, that branch routes to a physician instead of a chatbot. Wellness stays instant; medicine gets a name behind it.
Why the line matters →Keep your own build. Add the doctor.
Same connectors, same speed, same skipped $15k. Plus a physician on anything clinical, and your data where it belongs — with you. Where this stands today: the attestation engine is a working demonstration and the physician directory is live; a patient lab-review product is not live yet.