The pitch for clinical AI is usually about the consultation itself. In practice the consultation is the part that works. What does not work is everything packed around it, and that is where an agent earns its place.
The shape of it
An agent is not a window you open and ask. It runs alongside the clinic and produces work. The distinction shows up in what you find waiting rather than in what you can request.
- The note. Written from the consultation, in the format the clinic already uses, while you work rather than after everyone has gone home.
- The letters and the forms. Referrals, certificates, insurance forms, discharge summaries — the repetitive writing that fills the gaps between patients, drafted and waiting for a signature.
- The reading. A patient arrives with a report. It has been read, compared against their previous one, and what moved is surfaced — against the range the laboratory itself printed.
- The thread. What was ordered, what came back, what is outstanding, what you said last time.
Compression, not cleverness
The value is not that the system is smart. It is that ninety seconds of reading becomes ten. A clinician does not want a second opinion from software; they want the same information in a quarter of the time.
Where it stops
It drafts. The clinician composes and signs. Nothing leaves the room without that — not a note, not a letter, and above all not a prescription. And it does not interpret a named patient’s data to produce a diagnosis, because that is a regulated determination and the line is deliberate.
Counting is safe: “four of thirty-two values sit outside the laboratory’s printed range” is arithmetic. “This pattern suggests a condition” is not, and we do not cross it.
This article explains how our technology works. It is health information, not medical advice.