Blog
What the AI does, and what it will never do
Most of what Orla automates today is not AI at all, and saying so is the point. Here is exactly where the line sits, and what has to be true before it moves.
Every practice management system now says it has AI. Very few will tell you which parts of the product it actually touches, and that vagueness is doing a lot of work — because "AI-powered" over a feature that is a database query is a claim nobody can check.
So here is the line, drawn where it actually is today.
Most of what Orla automates is not AI, and that is a feature
The things clinics notice first are deterministic:
- Waitlist auto-fill matches waiting patients to a freed slot on stated rules — same service, the practitioner they named, the dates and the half of the day they said they could do — ranks them, and lets the first click win.
- Reminders and recalls send from templates on the clinic's schedule.
- Lapsed care plans and unconfirmed appointments surface because the system looks for plans past their date and appointments nobody has confirmed.
None of that involves a language model, and none of it should. You want the matching rule for a cancellation to be a rule: the same inputs give the same answer, you can read what it will do before it does it, and when a patient asks why they were offered a slot the answer is a rule rather than a probability.
Calling that "AI" would make it sound more impressive and less trustworthy at the same time.
What the actual AI layers will be, and their honest status
Three are designed into the system, and the console says plainly where each one stands rather than showing a mock of it:
- A text front desk that answers patient messages, books real slots within the clinic's policy and hands anything clinical straight to a human. Next phase. The inbox it plugs into is built.
- A scribe that, with the patient's consent, drafts a consultation note in the clinic's own template. Next phase. The clinical record it would write into — templates, signing, amendments, immutability — is live today.
- Voice, for the after-hours and mid-treatment phone. In pilot design, and deliberately last: it inherits every rule the text agent proves first.
We would rather tell you the scribe is not built than show you a video of one that is not.
Why it will not write clinical notes on its own
A note is a legal record of care. Signing one is a clinician saying "this is what I found and what I did", and that statement cannot be delegated to software, however good the draft.
So the boundary is: a draft stays a draft until a clinician signs it. Orla never signs. What would have to change for that to move is not model quality — it is professional regulation and indemnity, and neither of those is ours to decide.
Why consent is a property of the code, not a promise in a policy
Any vendor can write "we respect patient consent" in a privacy notice. The question is what happens when the code is asked to send something it should not.
In Orla, the consent a message requires is decided by the template, in one place, and checked on every send. That design choice has a specific consequence: a new message type added a year from now inherits the correct consent rule automatically, rather than depending on whoever adds it remembering to declare it. A send without consent is refused and the refusal is recorded.
When the AI layers arrive they send through that same path. They get no exemption, because the guard is not a policy the assistant is asked to follow — it is a gate the message has to pass.
What a clinic is agreeing to, in plain words
The rules the assistant will run under, as the console states them:
- It always says it is an assistant. It never pretends to be a person.
- Clinical questions go to your team verbatim. It never advises.
- Prices and policies come from your settings. It does not guess or invent.
- Drafts stay drafts. A clinician signs; Orla never does.
- Every action it takes is in the audit log, attributed to the assistant.
That last one is why the audit log has always distinguished a person from the system from an assistant from the API. The distinction was built before there was an assistant to record, because retrofitting accountability onto a system that already acts autonomously is how you end up unable to answer "who did this".
The short version
Today Orla automates your admin with rules you can read. Tomorrow it will answer your texts and draft your notes, under the constraints above. When it does, this page will say so — and until it does, this page will keep saying it does not.