Skip to content
be-human.ai
← Back to the journal

AI by industry

AI Agents for Plastic Surgery Offices: Consults and Photo Consent

Which consult requests and photo consent steps a plastic surgery office can delegate to an AI agent, and which must stay under surgeon control.

Be Human typographic cover with the title: AI Agents for Plastic Surgery Offices: Consults and Photo Consent
Conceptual illustration created with AI · Be Human

A plastic surgery office can safely hand an AI agent the routine parts of consult intake, the reminders that collect before-photos and consent forms, and the follow-up queue for patients who have not replied. Clinical judgment, candidacy decisions, anything about risk or outcome, and the final consent conversation stay with the surgeon and clinical staff. The practical line is simple: the agent moves information and reminders, while people decide what a patient is medically and legally agreeing to.

Key takeaways

  • Delegate intake logistics, scheduling and document reminders, and keep candidacy, risk and clinical answers with the surgeon.
  • Treat before-photo consent as a staged workflow with a named human owner at each step, not a single checkbox.
  • Give the agent narrow permissions: it can read a summary, send approved templates and log activity, but it cannot edit clinical notes, change consent status or promise outcomes.
  • Measure success by complete consult packets before the visit and faster human replies to escalated messages, not by message volume.
  • Start with one queue, one owner and a written escalation rule before adding channels or tools.

Where an agent fits in the consult pathway

A plastic surgery consult moves through a sequence that is mostly administrative before it becomes clinical. An inquiry arrives, a consult is booked, a questionnaire is completed, photographs are requested, consent is documented, and then the surgeon meets the patient. Each handoff is a place where information goes missing. The useful question for an operations leader is not whether an agent can talk to patients, but which handoffs are logistics that can be checked against a written rule, and which are judgments that belong to a clinician.

Be Human's first engagement defines one role, the memory and tools that role needs, the approval boundaries, and one observable success criterion. In a surgical office, a sensible first role is a consult coordinator with a narrow scope, not a general assistant that answers everything. A narrow role is easier to test, easier to audit and easier for staff to trust.

Consult requests an agent can handle

Most inbound messages before a consult fall into a small set of types. An agent can handle these safely when the practice has already written down the answers:

  • Logistics questions: office address, arrival time, parking, what to bring, and whether a companion or an interpreter is welcome, in English or Portuguese.
  • Booking requests that fit rules the surgeon set, such as consult length, the days new patients are seen, and how far ahead a slot can be held.
  • Pre-consult questionnaire collection, including noting blank fields and asking the patient to complete them.
  • Requests to reschedule, cancel or join a waitlist, recorded in the practice system through a tool the agent is permitted to use.

Some messages must reach a human at once. Any description of symptoms, a new medication, a previous complication, a pregnancy, or a question about whether the patient is a candidate should be routed without an answer from the agent. The agent can say that a staff member will reply and state the expected response window. It should not interpret the symptoms or suggest what the surgeon would decide.

Before-photo consent as a staged workflow

Before-photos are clinical records and personal images at the same time, so the consent process needs to be a sequence with an owner at each stage. A workable plan looks like this:

  • Step one: the agent sends the photography instructions approved by the surgeon, covering lighting, distance and which items to remove, in the language the patient chose.
  • Step two: the agent sends the privacy and image-use notice from the approved text and records that it was sent.
  • Step three: the patient uploads images to a restricted folder linked to the correct record. The agent checks that files arrived and are attached to the right patient, and does nothing with the content beyond that.
  • Step four: a staff member reviews the images for completeness and marks the packet as received. The agent cannot mark it as approved.
  • Step five: the clinician holds the consent conversation, in person or by video, and records it in the chart. Only the clinician's entry changes the consent status.
  • Step six: permission to use images for teaching or marketing is a separate, optional and revocable request. The agent never bundles it into the clinical consent message.

Health information and images are sensitive personal data in most jurisdictions. In Brazil they fall under the LGPD, which the ANPD supervises. Confirm with your legal adviser what the practice must record and how long images are kept, and let the agent inherit those rules rather than invent its own.

Illustrative scenario: a consult coordinator at a small practice finds that three patients uploaded photos last week, but none has a signed consent on file. The agent's daily queue shows the three names, the upload dates and the missing step. The coordinator asks the surgeon's assistant to call them. The agent sends only the generic missing-document message the surgeon approved, and it does not mention the procedure.

What stays under surgeon control

Some decisions should never be delegated, even partially. These stay with the surgeon and clinical staff:

  • Candidacy, surgical plan, technique and expected recovery.
  • Risk and complication discussions, including what to say when a patient asks how a procedure could go wrong.
  • Interpretation of before-photos, including whether an image shows a result the patient hopes for.
  • Any message about results, promises of outcome, or comparisons with other patients.
  • Post-operative symptoms, medication changes, wound questions and anything that might be urgent.
  • Revision decisions, exceptions to a signed plan, and any change to terms already agreed.

The agent should be built to refuse these topics by design, not by trusting that a prompt will keep it careful. Instructions help, but tool permissions do more of the work. The agent should not be able to change clinical notes, alter a consent status or send free-text messages about clinical content. OWASP's authorization guidance describes this least-privilege pattern, where each component receives only the access its task requires. Patient messages are also untrusted input. Someone may write text designed to override the agent's rules, so the system should treat incoming messages as data to classify, never as instructions to follow.

Follow-up queues and how Be Human runs them

The follow-up queue is where much of the value sits, and much of the risk too. It usually holds consults with no response, consent packets not yet returned, decisions pending after a consult, pre-operative checklists and scheduled check-ins after surgery. Each item needs an owner, a timer, an approved template and an escalation path. Post-operative check-in messages can be scheduled, but any reply containing a symptom term goes to on-call clinical staff immediately, and the agent does not triage it.

In a managed engagement, Be Human defines the consult role, the memory and tools it may use, the approval boundaries and the success criterion before anything goes live. Operations then cover monitoring of queue health, recurring evaluations against realistic patient messages, permission reviews when staff change roles, and incident response if a message goes out that should not have. Our notes on intake and recall work for a chiropractic practice describe a similar split between routine logistics and clinical questions.

Measuring whether the agent is doing its job

Measure what the office already tracks: how many consult packets are complete before the visit, how quickly a human answers an escalated message, how often staff correct an agent message before it goes out, and how many queue items sit past their timer. Avoid counting messages sent, because that rewards volume rather than correct handling. The NIST AI Risk Management Framework is a useful reference for deciding which risks to measure before launch and who reviews them.

Build a small evaluation set from real requests, anonymized and reviewed by staff, and include the hard cases: a patient asking whether they are a good candidate, a patient who mentions a fever after surgery, and a patient who wants to withdraw image permission. Run the agent against this set whenever its instructions, tools or templates change. Anthropic's guidance on building effective agents favors simple, composable patterns with clear tool boundaries over broad autonomy, which fits a practice that needs predictable behavior more than novelty.

If you want to find the first task worth handing to an agent in your practice, take the free Be Human AI diagnostic.

Frequently asked questions

Can an AI agent collect before-photos from patients? Yes, an agent can send approved photography instructions, remind patients about missing uploads and confirm that files are attached to the right record. It should not judge image content or mark consent as complete. A staff member reviews the packet, and the clinician holds the consent conversation.

Should the agent answer whether a patient is a candidate for surgery? No. Candidacy, risk and surgical planning are clinical judgments for the surgeon. The agent can explain office logistics, collect the questionnaire and route the question to a human with a stated response window, even when the question sounds simple.

How do we keep patient photos out of marketing by accident? Keep marketing permission as a separate, optional and revocable request that the agent never bundles with clinical consent. Store images in restricted folders, limit access to the roles that need them, and review those permissions whenever staff change. Confirm retention and use rules with your legal adviser, because health and image data carry specific obligations.

What should a first project cover? Start with one consult coordination role, a narrow queue such as incomplete consult packets, a written escalation rule for clinical words, and one success criterion the office can check every week. Add channels and tools only after the first queue runs cleanly. A first engagement should be small enough to observe and correct.

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.

Less busywork. More room for you.

Bring us the work you wish you had time for.

Book a call to try a personalized AI agent free for 7 days ↗Or take the free AI diagnostic →