Superintelligence Prep for Acupuncture Clinics: Intake and Recalls

For an acupuncture practice, an AI agent is most useful once intake, visit notes and recall rules are written down clearly enough that a person could follow them by hand. The agent then handles the repetitive coordination around those rules, such as preparing intake summaries, drafting recall messages and flagging overdue treatment plans, while the practitioner keeps every clinical decision. Structure these three workflows first, then bring in a scoped agent for one of them with an approval step and a plain success measure.

Key takeaways

• Write intake, visit notes and recalls as explicit rules before any software acts on them.
• Keep diagnosis, treatment choice and clinical narrative with the licensed practitioner.
• Start an agent with one bounded role, a named memory source and a human approval step for every outbound message.
• Measure the workflow with counts you can check each month, such as due recalls, reviewed messages and returns within the planned window.
• Widen permissions only after the first workflow produces consistent output under review for a sustained period.

Why intake is the first workflow to structure

Intake is where an acupuncture clinic collects the information that every later decision depends on: why the patient came in, relevant medical history, current medications, previous treatment, and the consent forms that must be signed before the first session. In many clinics this information is split across paper sheets, a practice management form and notes kept at the front desk. Before any agent touches it, the clinic should agree on one current version of the intake form and one place where completed intakes are stored.

The useful distinction is between collecting and interpreting. Collecting is administrative: confirming that every required field is complete, that the consent form is signed, that contact details are current, and that the practitioner has the packet before the appointment. Interpreting belongs to the licensed practitioner, who decides whether a presentation is suitable for treatment, when to refer and what to avoid. An agent can support the first group. It should not make the second group's decisions, and the intake design should make that split visible on paper.

A practical first step is to write three lists: the fields that must be complete before the first visit, the fields the practitioner reviews in the room, and the fields that can be updated later. That document is also the rule set an agent would follow. Clinics in adjacent fields have worked through the same questions, and the approach in the AI agent for a chiropractic practice covering intake and visit recalls shows how these field lists translate into a working workflow.

What a visit note should contain before an agent reads it

The visit note is the clinical record of what happened in the room, and it is the source most likely to be misread if an agent is given too much freedom. The practitioner should write and sign the note. What an operations agent may need is a structured summary of agreed fields: the date, the visit type, the plan the practitioner set, the patient's reported response, and the follow-up interval the practitioner recommended. Those fields are factual enough to support scheduling and recall without the agent writing any clinical content.

A visit note template that supports recall usually includes:

• Visit date and visit type, such as initial, follow-up or re-evaluation
• Treatment plan agreed with the patient and written by the practitioner
• Recommended interval until the next visit, in the practitioner's own words
• Patient-reported response, recorded as the patient stated it
• Contraindications or reasons to hold a treatment, flagged by the practitioner
• Communication preferences and consent, confirmed or updated at the visit

Keep the free-text clinical narrative in the chart rather than in the agent's working memory unless a specific task requires it. Practices that plan care around structured plans face the same split between plan and narrative, which is discussed in AI for dietitian practices covering intake forms, plans and follow-up.

Designing recall rules a clinic can explain to patients

A recall is a reminder that a patient should return. In an acupuncture practice it usually depends on a treatment plan with a stated interval. The rule should be simple enough that the front desk, the practitioner and the patient all describe it the same way. A recall nobody can explain will be overridden by staff, and an agent that follows an unexplained rule will produce messages no one trusts.

A workable rule set follows a short sequence:

• Define which patients are on an active treatment plan and which have finished it.
• Write the interval rule for each plan type, taken from the practitioner's recommendation.
• Define the gap after the last visit that triggers a reactivation message, and name the person who approves its wording.
• Record each patient's consent to messages, preferred channel and quiet hours.
• Set an opt-out path that stops messages at once and is logged.
• Decide which messages are sent automatically and which are drafted for staff approval.
• Review the rules monthly with the practitioner and the person who answers the phone.

Illustrative scenario: a patient completed a treatment plan set by the practitioner and has not returned within the recommended window. The rule says the reactivation message is drafted into a staff review queue, not sent. A front desk employee reads the draft, adds a line about the clinic's current schedule, and approves it. The message goes out through the patient's preferred channel, and the approval is recorded with the patient's file.

Setting memory, tools and permissions for a scoped agent

A scoped agent needs three things defined before it runs: the memory it may read, the tools it may use, and the actions that require a person's approval. Memory should be limited to approved sources, such as the current intake template, the recall rules, the schedule and the message log. Tools should be narrow. A reasonable first set is read access to the schedule, read access to the recall list, and the ability to place a draft message in a review queue. Sending messages, editing charts and changing a treatment plan should stay outside the agent's permissions.

Permissions should follow least privilege, so the agent holds only the access its role needs and every grant can be reviewed and removed. The OWASP authorization guidance describes this principle for application access in general. Patient messages also deserve caution. Text written by a patient can contain instructions that try to change how an automated assistant behaves, so the agent should treat incoming message content as material to summarize, never as commands. The OWASP Top 10 for LLM Applications lists this kind of input manipulation among its main risks.

Data protection rules apply from the first day. Depending on where the clinic operates, the relevant rules may be HIPAA in the United States or the LGPD in Brazil, and the Brazilian national data protection authority, ANPD, publishes guidance on the LGPD. Clinics that handle sensitive health information can compare their boundaries with the AI agents for behavioral health clinics and the limits of intake and scheduling before widening any role. For a broader frame on documenting risks and controls, the NIST AI Risk Management Framework is a useful reference.

When to bring in a scoped agent and when to wait

Bringing in an agent is a decision about readiness, not enthusiasm. The clinic is ready when the workflow has a written rule, a named owner, a reviewer who can check output quickly, and a way to switch the agent off without disrupting the front desk. If any of these is missing, the work stays with staff for now, and the missing piece becomes the first task.

Use this guide to decide:

• Bring in an agent when the intake template is final and stored in one place.
• Bring in an agent when the recall rule can be explained to a patient in two sentences.
• Bring in an agent when one person owns the workflow and answers for its errors.
• Bring in an agent when a reviewer can approve each output in a short, fixed step.
• Wait when the practitioner and the front desk describe the recall policy differently.
• Wait when consent and channel preferences are kept in several places that do not match.
• Wait when no one has decided who is alerted when the agent fails or is uncertain.

Superintelligence preparation at this scale is mostly ordinary discipline. Clean intake fields, explicit clinical boundaries and logged approvals make a practice ready for more capable tools later, and they are useful even if no agent is ever deployed.

How Be Human runs this in a managed setup

Be Human starts with one useful role, which in an acupuncture practice is often the recall coordinator or the intake reviewer. The engagement defines the memory and tools that role needs, the actions that require approval, and an observable success criterion, such as the number of due recalls that received a reviewed message each week. Managed operations then cover monitoring, recurring evaluations of the agent's drafts against the clinic's written rules, permission reviews, integration maintenance and incident response when a message or record goes wrong. The practitioner keeps clinical authority throughout, and any change to the agent's scope is made deliberately and recorded. The service is remote and available in English and Portuguese.

Measuring recalls without guessing

A recall workflow is working when someone can answer three questions from the records without asking the agent: how many patients were due, how many received a reviewed message, and how many returned within the window the practitioner set. Track these as plain counts and review them with the practitioner each month. If a count moves unexpectedly, check the rule and the message log before changing the agent. Evaluation practice in the OpenAI evals guide describes the same habit of comparing outputs against fixed expectations before a workflow is widened.

If you are not sure which of these workflows should go first, take the free Be Human AI diagnostic to find the first task worth handing to an agent.

Frequently asked questions

Can an AI agent write acupuncture visit notes? It should not write the clinical record. The practitioner writes and signs the note, and an agent can at most read a structured summary of agreed fields to prepare scheduling or recall tasks. Keep the clinical narrative in the chart and limit the agent's permissions to what the workflow actually needs.

Which workflow should an acupuncture clinic automate first? Usually the one with a written rule and a single owner, which is often recall rather than intake. Intake depends on a stable form and a settled consent process, while recall reminders can be drafted into a review queue as soon as the interval rule is agreed. Check that the rule can be explained to a patient in a sentence before starting.

Does a recall agent need to send messages on its own? Not at the start. Drafting into a staff review queue lets a person approve each message, and the approval step should be relaxed only after evaluations show consistent output over a sustained period. Any automatic sending should be a separate permission decision, recorded in the operating rules and reviewed on a schedule.

How does this relate to superintelligence? The term describes capability that may grow, but the preparation is the same whatever the tools become: clear records, explicit boundaries and logged approvals. A clinic that does this work now can add more capable tools later without rebuilding its memory or its permissions.

Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.