The AI Doctor Rounds at Three in the Morning with Chanakya Toleti, President, Psynergy Health
At three in the morning, something is rounding on every patient in the program. It reads the overnight vitals and the new labs. It picks up the medication change entered at eleven and the note the covering physician dictated after it. It compares each patient against how that patient looked last week rather than against a population average. By the time the day team logs in, the list is waiting: these six drifted, this one has not filled a prescription in nine days, this one's weight is up four pounds, call her first. Call it the AI doctor. It rounds on every chart, every night, and it never reaches the end of a shift.
What it does not do is take care of anyone. The industry keeps getting this backwards, which is why a decade of software purchases has not closed a single capacity gap. Reading a chart is not rounding. The AI doctor produces a list. Somebody with a license and a phone still has to work it.
What the AI doctor is good at
Completeness, mostly. A human team responsible for tens of thousands of chronic-disease patients cannot look at every record every day, so it looks at the ones that surface: the patient who called, the patient with an appointment tomorrow, the patient a nurse happened to remember. The AI doctor looks at all of them, which changes who gets found. It is also better than a person at the work that was never the point of a nursing license, such as pulling the relevant signal out of a long record, drafting the outreach, and noticing that a value moved in a direction it should not have.
What it is not good at
Everything after the list. Whether a twelve-point drift matters for this particular patient, with this history, on this combination of medications. Whether the hesitation in someone's voice means they do not understand the plan or cannot afford it. Whether to adjust the dose or bring them in. And accountability, which cannot be assigned to software. The AI doctor sorts the queue. A nurse practitioner owns the outcome.
There is a harder question underneath that one, which is what happens when the list is wrong. The AI doctor will miss somebody. A patient will deteriorate in a pattern it was not built to recognize, or a data feed will go quiet and nobody will notice the silence. In a software purchase, that failure lands on the health system, which bought a tool and accepted the terms. In a clinical service, it lands on the clinicians who carry the panel, along with the obligation to find out why and change the process. That difference is written into a contract or it does not exist, and it is what separates a vendor from a partner.
"Everyone wants to argue about whether AI replaces the physician," says Chanakya Toleti, president of Psynergy Health. "The better question is what the physician stops doing. Our AI rounds on every chart overnight, so the clinical team starts the day on the sixth patient instead of the first. No judgment got automated. The queue got sorted."
Why this is a services question
Here is the trap waiting for any health system that buys the AI doctor as a product. The output of good clinical AI is more identified need, which is only useful to an organization with the staff to meet it. Hand a short-staffed unit a longer list of patients who need a call today and the technology has made the day worse. This is why systems are increasingly buying clinical services instead of clinical software. The AI and the clinical team arrive together, one sorting the work and the other doing it, with the outcome owned on the partner's side of the line rather than the buyer's. AI-native clinical services means exactly this: the software and the clinicians are built as one thing, because separately neither of them finishes the job.
The AI doctor will round all night and cost almost nothing to run. Someone still has to pick up the phone at eight in the morning, and that is the thing American healthcare is actually short on.






Comment