Clinical AI's early use is preparing the visit while keeping the doctor in control

Original research record

We connect the OTR data on AMIE's real-patient study with the wider move towards agents that complete work. At one clinic, 98 of 114 enrolled patients completed both the AI interaction and their appointment. Physicians monitored the conversations. No encounter required a predefined safety-stop intervention; one hallucination and five clarification cases were recorded. Our synthesis is that collecting history and preparing a clinician may be the nearer deployment path than independent medical decisions.

We would measure clinician preparation time, useful information, corrections, patient trust and adverse events. A generated summary should preserve uncertainty and support a doctor's review.

Larger independent trials showing safe care outcomes with less supervision would support broader autonomy. Poor patient trust or increased review work would weaken the support-workflow case.

We see a credible design path in supervised preparation. This is a technology deployment signal; we do not turn a feasibility study into personal medical advice.

Bayesian · what changed the estimate? We assign a subjective 65% chance that a one-week document review will find supervision and clinician review central to this deployment, rather than evidence for unsupervised care. This probability concerns the stated evidence boundary. we have not fitted a prior or evidence likelihoods, so this is not a computed posterior. de Finetti · are the comparisons fair? we treat repeated OTR editions as one underlying source when they reuse a report; our judgement must not gain confidence just from repetition. Kolmogorov · do the numbers agree? for the stated test event, the probability of it occurring and not occurring sums to 100%. None of these checks establishes historical calibration.

The OTR report and original hospital page preserve the denominator and monitoring arrangement. The project site lists further prospective studies, which are not completed outcome results.

Not applicable: this is a technology synthesis, not an investible or price forecast.

The small, single-centre sample cannot establish rare-harm rates. Strong conversation ratings do not show reduced workload or better health.

Not applicable: no earlier market direction is assigned. We state the evidence dates and the condition that would change the technology view.

We separate encouraging feasibility from proof that a chatbot can replace a clinician. The reported diagnostic match also depends on how many possible diagnoses were included.

We have not verified completed, matching technology forecasts in the OTR records inspected for this signal. The percentage is an uncalibrated judgement and is not a historical hit rate.

How likely is this call? · Three frameworks

Bayesian · what changed the estimate? We assign a subjective 65% chance that a one-week document review will find supervision and clinician review central to this deployment, rather than evidence for unsupervised care. This probability concerns the stated evidence boundary. we have not fitted a prior or evidence likelihoods, so this is not a computed posterior. de Finetti · are the comparisons fair? we treat repeated OTR editions as one underlying source when they reuse a report; our judgement must not gain confidence just from repetition. Kolmogorov · do the numbers agree? for the stated test event, the probability of it occurring and not occurring sums to 100%. None of these checks establishes historical calibration.

Bayesian · what changed the estimate?

We assign a subjective 65% chance that a one-week document review will find supervision and clinician review central to this deployment, rather than evidence for unsupervised care. This probability concerns the stated evidence boundary. we have not fitted a prior or evidence likelihoods, so this is not a computed posterior.

de Finetti · are the comparisons fair?

we treat repeated OTR editions as one underlying source when they reuse a report; our judgement must not gain confidence just from repetition.

Kolmogorov · do the numbers agree?

for the stated test event, the probability of it occurring and not occurring sums to 100%. None of these checks establishes historical calibration.

Evidence behind this view

The OTR report and original hospital page preserve the denominator and monitoring arrangement. The project site lists further prospective studies, which are not completed outcome results.

What past price moves show

Not applicable: this is a technology synthesis, not an investible or price forecast.

Evidence against this view

The small, single-centre sample cannot establish rare-harm rates. Strong conversation ratings do not show reduced workload or better health.

Why the direction changed

Not applicable: no earlier market direction is assigned. We state the evidence dates and the condition that would change the technology view.

What people say vs the facts

We separate encouraging feasibility from proof that a chatbot can replace a clinician. The reported diagnostic match also depends on how many possible diagnoses were included.

How past forecasts performed

We have not verified completed, matching technology forecasts in the OTR records inspected for this signal. The percentage is an uncalibrated judgement and is not a historical hit rate.