Why the AI stays behind the clinician

An AI scribe helps draft the record of a visit. The clinician still has to check that record, make the diagnosis, and agree on a plan with the patient. That is why I want the AI behind the clinician: useful enough to help with the chart, with a clear boundary around who makes decisions about care.

For me, that boundary has three parts: check the note, question the recommendation, and keep responsibility with the clinician.

Why does the AI stay behind the clinician?

A useful tool leaves the clinician able to disagree with it. A visit involves choices about what to ask, which details matter, and what to do when the history does not fit the first explanation. I want software to support that work while leaving the clinician room to investigate and change course.

Consider a hypothetical note that turns "the patient wondered whether the medicine caused the symptom" into "the medicine caused the symptom." The wording is tidy. The meaning has changed. A readable draft still needs someone who knows what happened in the visit to check it.

What should an AI scribe do?

The draft should be easy to inspect and correct. My test for a scribe starts with whether a clinician can tell what it captured, what it missed, and what it inferred. A polished paragraph is useful only if it represents the encounter faithfully.

Part of the visit	Boundary I want the software to respect
Documenting the history	Help draft the record; leave the clinician to verify the patient's account.
Expressing uncertainty	Preserve what is suspected, reported, or still unknown.
Recording a plan	Capture the clinician's decisions without silently adding a new one.
Signing the note	Leave an editable draft for the clinician to review.

These are the questions I would use to evaluate a tool, not a claim that software can check its own work. Ask to see how it handles a correction and a missing detail before judging it by the best-looking example.

How is clinical decision support different from an AI scribe?

Documentation records the encounter; clinical decision support offers information or recommendations that may influence care. Clinical decision support, or CDS, can help a health professional consider prevention, diagnosis, or treatment. That makes the basis for a recommendation worth examining separately from the quality of a generated note.

As checked on October 12, 2026, 21 U.S.C. § 360j(o)(1)(E) excludes certain CDS functions from the definition of a medical device when all its criteria are met. One criterion is that a health professional can independently review the basis for the recommendation rather than rely primarily on it. There are other criteria, including limits involving medical images and certain signals. Human review alone does not settle a product's regulatory status.

The FDA's January 2026 Clinical Decision Support Software guidance explains how the agency interprets those criteria. This essay describes my design priorities; it is not legal advice or a regulatory classification of a Vye product.

Before relying on a recommendation, I would want a tool evaluation to answer:

1. What patient information and evidence produced it?
2. What assumptions or missing information could change it?
3. Can the clinician inspect that basis and choose a different course?

What does this mean for Vye Health?

I am not a clinician. The clinicians are. My job is to keep the tools from moving into their chair.

Charting, labs, and telehealth in one place, plus an AI scribe.

That is the scope to discuss when you apply to Vye Health. If you are still planning the practice itself, start with the private-practice launch guide or the telehealth launch checklist.