Clinical Inbox & Messaging AI

AI that works the clinical inbox: drafting replies to patient portal messages, triaging inbound messages by urgency, releasing and explaining results, and disposing of refill requests and forms without a clinician handling each one. The category is distinct from ambient documentation, which writes a new note from a live encounter, and from patient voice agents, which do a comparable job synchronously over the phone. Three properties make the asynchronous case different enough to evaluate on its own terms. The failure is a delayed response to an urgent complaint rather than a dropped call. The drafted reply goes out under a named clinician's identity, so the attribution and licensure surface belongs to the person who signs it rather than to the vendor. And the work happens inside the EHR inbox itself, which makes integration depth a precondition rather than a differentiator. The decisive evaluation question is mistriage: what the system does when an urgent clinical signal arrives inside a routine looking message, and what proportion of generated drafts clinicians send unedited. Buyers should treat time saved with caution as a headline metric, since published evaluations of drafted portal replies have tended to find the effect in cognitive burden rather than in minutes, and should establish in advance who is accountable for a reply that a model drafted and a clinician signed.