Superintelligent Agents for Oral Surgery Offices: Referrals and Recalls
How an oral surgery office can use AI agents for referral intake and post-op recalls now, which steps need human review, and how to measure results.

An AI agent can help an oral surgery office take in referrals, check that each referral packet is complete, and keep recall and post-op follow-up tasks moving under rules the office sets. The agent prepares, routes and reminds, while surgeons and clinical staff make every clinical decision and approve any message with clinical content. Built carefully today, the same governed records and permissions can later support more capable agents without rebuilding the office's process.
Key takeaways
- Referral intake is the most practical first workflow because it is high volume, text-heavy and easy to check against a required-information list.
- Post-op follow-up needs surgeon-approved scripts and escalation rules before any agent message reaches a patient.
- The value compounds in the referral record: what arrived, what was missing, what was promised and what happened after surgery.
- Every agent action needs a named permission, a log entry and a human owner for exceptions.
- Start with one role and one success criterion, then widen scope only after the office can read the logs and trust how exceptions were handled.
What an agent would actually do in the office
Oral surgery referrals arrive through several doors: a fax from a general dentist, a portal upload, an email with attached imaging, or a phone call that becomes a handwritten note. Each arrival carries the same problem. Someone has to find the patient, confirm identity, read the clinical summary, identify missing items such as imaging, medical history, medication list and anticoagulant status, and decide who on the team needs the packet next. That work is repetitive and mostly reading and sorting, which is why it is a good starting point for an agent.
Recall and post-op follow-up have the same shape. After a procedure, the office follows a protocol the surgeon has written: when to call, what to ask, when to send care instructions and when to schedule a check. Most of the time the work is making sure a scheduled message goes out and that the response is logged. The clinical judgment stays with the team.
What the agent should not do is just as specific. It should not diagnose, interpret imaging, recommend medication changes or decide whether a symptom is urgent. When a patient message contains language that suggests a complication, the agent's only job is to stop its own flow and route the case to a named clinical person with the original message attached.
Why the referral record is the real asset
As agents become more capable, the limiting factor in most offices stops being whether a model can write a reply and becomes whether the office can see what happened in each case. A referral that moves from intake to consult to surgery to a post-op call creates a chain of facts. If those facts live in a shared inbox, a spreadsheet and three people's memories, no agent can use them reliably. If they live in one governed record, with timestamps and sources, an agent can answer the questions that matter: was the medication list ever received, who approved the post-op script for this procedure, and which follow-up was skipped.
This is what the superintelligence lens means in practice. The point is not that a future system will know everything on its own. The point is that an office that builds a clear, permissioned case record now gives any future agent something trustworthy to work from. Retrieving approved documents, such as the surgeon's protocol or the office's referral checklist, is usually safer than asking a model to remember policy. Those documents still need a named owner who keeps them current.
Other referral-heavy practices face similar intake questions, as described in the physical therapy referral guide.
A referral intake plan to build now
The first version should be deliberately plain. Build it in this order and do not skip the review steps.
- Step 1: Write the required-information list with the surgeon and the coordinator. Include identity fields, referring provider, reason for referral, imaging received, medical history, medication and allergy list, and any anticoagulant or sedation question the surgeon wants flagged.
- Step 2: Give the agent read access to the one inbound channel the office already uses, such as the referral email address or the portal, and nothing else.
- Step 3: Have the agent extract the fields into the case record and mark each required item as received, missing or unclear, with a link to the source document.
- Step 4: Route any packet with a missing or flagged item to the referral coordinator as a draft request to the referring office. The coordinator approves before anything is sent.
- Step 5: Hold complete packets in a queue for the surgeon's review, with a short summary that lists what the packet contains and what it lacks. The agent does not offer a consult time until the coordinator confirms the packet is ready.
- Step 6: Log every action with a time, the actor and the source. Review the log weekly for the first month.
Be Human approaches this as a managed service. A first engagement defines one useful role for the office, the memory and tools that role needs, the approval boundaries and one observable success criterion. Managed operations can then cover monitoring, recurring evaluations, integration maintenance, permission review and controlled changes to the agent, while the office's surgeon and coordinator remain the people who approve anything clinical.
Recalls and post-op follow-up without clinical drift
Post-op follow-up is where small wording errors carry the most risk, so the rules should be written before the agent sends anything. Work with the surgeon to produce a short library of approved message templates for each common procedure, covering when the first call is due, what the patient is asked and which care instructions are standard. The agent may choose and personalize only from that library. It should not write new clinical content.
Recall work has a different risk profile. A missed recall is usually an administrative failure rather than a clinical one, so the agent can own the reminder schedule, the outreach attempts and the logging of responses. A patient who replies with a question about medication, bleeding, swelling or fever should be routed to a clinical person immediately, not answered by the agent. The same pattern appears in chiropractic intake and recall work, where the agent keeps the schedule moving and people handle anything that needs judgment.
Inbound text deserves caution too. A referral note or a patient reply can contain instructions that look like commands to an automated system. Treat that text as information about a patient, never as a request to change a schedule or to send a message outside the approved flow. Guidance on prompt injection prevention treats untrusted input in this way, and it is a sensible design rule for any office workflow.
Patients also reach offices after hours. Write down exactly what the agent may tell them: the office hours, the on-call number and the standard instruction to call for urgent symptoms. Anything beyond that goes to a person.
Approval boundaries and permissions
Permissions should be written as a plain decision guide that the office manager can read in two minutes. Sort every task into three groups: what the agent may do alone, what it may draft for approval, and what it may never do. Authorization guidance is a useful reference for the principle that each action should be checked against the permissions of the specific role involved, not against the agent's general access.
- The agent may act alone: read the designated inbox, create case records, mark missing items, set reminder times from approved templates and log outreach attempts.
- The agent may draft for approval: requests to referring offices, scheduling confirmations and any patient message outside the templates.
- The agent may never: give clinical advice, interpret imaging, change a surgeon's protocol, alter a medication list, cancel a surgical date or send a message containing a clinical claim.
Review these boundaries every month. Permission drift is common, because a shortcut added during a busy week tends to stay. The NIST AI Risk Management Framework offers a general structure for mapping and managing these risks, which helps when an office wants a written process its staff can audit.
Illustrative scenario: a referral with a gap
This scenario is illustrative and does not describe a real office or patient. A general dentist sends a referral for the extraction of a lower wisdom tooth. The packet includes the clinical note and a panoramic image, but the medication list is blank. The agent marks the medication field as missing, drafts a request to the referring office and places the packet in the coordinator's queue with a summary stating that the consult cannot be booked until the list arrives. The coordinator approves the request. When the list comes back, the agent updates the record, flags one anticoagulant for the surgeon's review and offers only the consult slots the coordinator has approved. After the procedure, the agent schedules the post-op call using the surgeon's template. The patient reports more swelling than the template describes. The agent does not answer the question. It routes the full message to the on-call nurse, logs the time and notifies the coordinator.
Each step leaves a trace that a manager can read later. That trace is what makes the next improvement safe.
How to measure whether it is working
Choose a success criterion that the office can observe without trusting the agent's own summary. Good candidates include the time from referral receipt to a complete or flagged packet, the number of referrals that reach the surgeon without a required item, and the share of recall attempts logged with a response or an escalation. Track each one from the source records, not from the agent's self-report.
Telemetry helps here. Traces, logs and metrics are most useful when they are tied together, so that a single case can be followed from arrival to the post-op call with each step visible to the person accountable for it. That is the practical meaning of observability for an office that does not have an engineering team.
Evaluate the agent on a set of past referrals, with patient identifiers removed, before you widen its role. Include the awkward cases: a referral with conflicting medication lists, a referral written in a hurry, and a post-op message that mentions a symptom in passing. If the agent handles those by routing rather than resolving, it is ready for a wider role. If it does not, change the rules before changing anything else. Share the results with the surgeon and coordinator, and keep the evaluation records with the logs. If you are unsure which of these tasks should come first in your office, the free Be Human AI diagnostic can help you find the first task worth handing to an agent.
Frequently asked questions
Can an AI agent book oral surgery consultations by itself? It can propose time slots and draft confirmations, but in the approach described here a coordinator approves any booking, especially when the referral packet has a missing or flagged item. Automatic booking can be considered later, once the office has logs and a track record of exceptions handled correctly. Until then, the agent's role is to prepare the work so a person can approve it quickly.
Is it safe for an agent to send post-op messages to patients? It is reasonable only when every message comes from a surgeon-approved template and the agent cannot add clinical content. Any reply that mentions a symptom, a medication or a concern must go to a clinical person without an automated answer. Review the message log regularly to confirm that the boundary is holding.
What should a referral packet include before the office schedules a consult? Most offices need the referring provider, patient identity, reason for referral, relevant imaging, medical history, a current medication and allergy list, and any anticoagulant or sedation questions the surgeon wants flagged. The exact list belongs to the surgeon and coordinator, not to the agent. The agent's job is to check each packet against that list and report what is missing.
Where should a surgery office start if it has never used AI agents? Start with referral intake on one channel, because the work is high volume, text-heavy and easy to check. Define one success criterion, such as how quickly a packet is marked complete or flagged, and set a weekly log review for the first month. Expand the role only when the office can explain what the agent did in recent cases and why.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


