Superintelligent Agents for Pediatric Practices: Well-Visit Recalls
An AI agent can build well-visit recall lists and draft reminders for a pediatric office, while immunization and clinical questions go to clinicians.

An AI agent for a pediatric office can build well-visit recall lists from the practice's own scheduling and visit records, then draft reminder messages that staff review before anything is sent. It applies recall rules that clinicians have approved, and it stops at the boundary of clinical judgment: immunization questions, symptoms, developmental concerns and any exception go to a named clinician queue. The result is a steady follow-up process that the front desk does not have to rebuild every week.
Key takeaways
- Recall lists are a data task. The agent applies approved rules, and clinicians keep ownership of what a child needs.
- Reminder drafts should come from approved templates, with human review wherever your policy requires it.
- Immunization, symptom and developmental questions belong in a clinician queue with a clear response expectation.
- Every list, draft and routing decision leaves a record that a staff member can audit.
- Start with one recall type and one measurable outcome before adding anything else.
What a well-visit recall agent actually does
A well-visit recall is a reminder that a child is due or overdue for a preventive visit. Today this work is often done by hand: someone exports a list, checks last visit dates, writes messages and tracks who replied. An agent can take over the repetitive parts of that loop. It reads the appointment history it is allowed to see, compares each patient against the recall rules your clinicians have written down, and produces a list grouped by the action needed, such as schedule, call back or no action.
The agent then drafts the outreach. A good draft is short, names the practice, states the reason for the visit in plain language, and offers the ways to book. It does not explain what a vaccine does or suggest that a child is behind on care in a way that sounds like a medical verdict. Those choices belong in the approved template, written by the people who own the clinical voice of the practice.
Finally, the agent tracks what happened after the draft. If a family replies, books, declines or goes silent, that state is recorded so the next recall batch does not repeat the same message to the same family without a reason. The agent is useful because it keeps this memory consistent, not because it makes decisions about care.
Which messages belong to clinicians
The most important design decision is the routing rule. A recall agent should treat anything that asks for a medical judgment as out of scope. In practice, that means a short list of categories is always routed to a clinician or a trained staff member, and the agent says so plainly to the family.
- Questions about whether a vaccine is due, delayed, contraindicated or safe for a particular child
- Questions about catch-up schedules, missed doses or reactions after a vaccine
- Any mention of symptoms, injuries, fever, breathing, feeding or medication
- Developmental, behavioral or school concerns that a family raises in a reply
- Requests to change a recall rule, skip a visit or document an exception
- Messages from a parent or guardian whose authority to act for the child is unclear
The agent's job at this point is narrow. It acknowledges the message, does not answer the clinical part, creates a queue item with the original text and the child's recall status, and states the expected response path. The clinician decides what happens next. This keeps the agent's replies within what the practice has approved and gives clinicians a clean record of what was asked.
A step-by-step plan for the first recall workflow
A first recall workflow should be small enough to check in a week. The following sequence is a practical starting point for an office that already has a scheduling system and a written recall policy.
- Step 1: Have the medical lead write or confirm the recall rules in plain terms, including which visit types are tracked and which exceptions exist.
- Step 2: Define the data the agent may read, such as appointment history and contact preferences, and exclude everything it does not need.
- Step 3: Create two or three approved message templates in the practice's voice, with fixed placeholders for name, visit type and booking options.
- Step 4: Set the routing list from the previous section and assign an owner and a response expectation to each clinician queue.
- Step 5: Run the agent in draft-only mode for a limited batch, and have a staff member compare its lists against a manual check.
- Step 6: Review every edited draft and every misrouted message, then adjust templates and rules before widening scope.
- Step 7: Agree on the point at which sending without individual review is allowed, if the practice decides to allow it at all, and write that decision down.
Keep the first version boring. A workflow that correctly lists, drafts and routes a single recall type is worth more than a broad system that nobody fully trusts.
Illustrative scenario: one Monday recall batch
Illustrative scenario: A small pediatric office has one front desk coordinator and two clinicians. On Monday morning, the recall agent produces a list of children with visits due in the next several weeks and a shorter list of overdue visits. Each entry shows the last visit type, the approved template, and the booking options.
Staff review the overdue list first and approve drafts for families with no open messages. Two replies come in during the afternoon. One family asks whether a shot can be given later than the schedule suggests. The agent acknowledges the message, does not answer the question, and places it in the immunization queue with the child's recall record attached. The other reply asks to move an appointment, which the agent handles within the approved booking options and logs. By Friday, the clinicians have a short queue of questions they can answer in one sitting, and the coordinator has not had to reconstruct the week from memory.
How to set permissions and protect family data
A recall agent touches sensitive information about children and their guardians, so access should be narrower than the staff member who might use it. Use least privilege: the agent should read only the fields required for recall, write only to draft and queue records, and have no ability to change clinical notes, vaccine records or billing. Authorization rules of this kind are described in the [OWASP Authorization Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Authorization_Cheat_Sheet.html), and they translate well to a clinic: each tool the agent can call should have an explicit scope, and the scopes should be reviewed on a schedule.
Messages from families are also an input the agent should not trust blindly. A reply might contain text that tries to change the agent's instructions, whether by accident or on purpose. The guidance in the [OWASP prompt injection prevention cheat sheet](https://cheatsheetseries.owasp.org/cheatsheets/LLM_Prompt_Injection_Prevention_Cheat_Sheet.html) recommends treating external content as data, not as instructions, and keeping the agent's actions constrained by fixed rules. In a pediatric setting, that means an incoming message can never cause the agent to send a clinical statement, change a recall status or contact someone outside the approved list.
Finally, check the legal obligations that apply where the practice operates. Privacy rules for health and family information differ by country and state, and the practice's compliance lead should review the data flows, retention and vendor terms before the agent goes live. For Brazilian practices, the [ANPD guidance portal](https://www.gov.br/anpd/pt-br) is the official starting point for personal data rules.
How Be Human runs this as a managed service
Be Human designs, implements and operates personalized AI infrastructure for companies, including agents that keep their identity, memory, channels, tools, permissions and operating context under human control. A first engagement defines one useful role, here the recall role, along with the memory and tools it needs, its approval boundaries and an observable success criterion. Managed operations can then include monitoring, recurring evaluations, integration maintenance, permission review, incident response and controlled improvements to templates and routing rules. The clinical content, the recall policy and the clinician queues remain the practice's responsibility, and the managed service keeps the agent inside them.
How to tell whether the recall agent is working
A recall agent should be judged on a few observable criteria that the practice chooses before launch: how often staff edit drafts, how many items are routed correctly on first review, how long a clinician queue item waits, and whether any message went out that should have been held. Write these down as a baseline and review them on a fixed schedule. Evaluation methods for agents, including repeatable test cases and recorded outcomes, are discussed in the [Anthropic guide to demystifying evals for AI agents](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents). Keep a standing test set of tricky messages, such as a question about a delayed shot phrased as a booking request, and rerun it after every change. If your practice also runs other patient-facing workflows, the intake and visit recall approach for a chiropractic practice and the recall and aligner tracking workflow for orthodontists show how similar routing rules are built in adjacent clinical settings. If you are deciding where to begin, 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 decide which vaccines a child needs? No. The agent applies recall rules that clinicians have written and approved, and it routes any question about immunization content, timing, eligibility or reactions to a clinician queue. Its reply to the family acknowledges the question without giving a clinical answer.
Does every reminder need a staff member to approve it first? That depends on your policy and your local rules. Many practices start with human review of every draft, then decide later, in writing, whether certain approved templates can be sent without individual review. The agent should follow whichever rule you choose.
What data does the agent need access to? Only the fields needed for recall, such as the child's last preventive visit type and date, the contact preferences of the guardian and the available booking options. It should not have access to clinical notes, billing or unrelated records, and its permissions should be reviewed on a regular schedule.
How long does it take to set up a first recall workflow? The time depends on how clean the scheduling data is and how quickly the medical lead can confirm the recall rules. A useful first version is one recall type, a few approved templates and clear routing, and it should be tested in draft-only mode before any wider use.
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.


