Superintelligence Prep for Martial Arts Schools: Belt Tests and Renewals
Keep belt-test, membership and renewal records clean so an AI agent can send reminders without losing student history, with clear approval limits for staff.

An AI agent for a martial arts school is most useful when it can answer one question reliably: what does this student need next, and when? That means knowing each student's current rank, the test they are preparing for, their membership terms and their renewal date, all in one record that a staff member would also trust. Getting the school's memory into that shape is the real work, and it comes before any reminder is sent.
Key takeaways
- An agent can only remind a student as well as the record it reads, so rank, membership and renewal dates each need one owner.
- Begin with reminders that carry little risk, such as test-date notices and renewal windows, and keep pricing, cancellations and refunds with staff.
- Give the agent read access to the record and let it write only to a draft queue that a person approves.
- Keep a dated change log for every promotion, freeze and plan change so the history survives staff turnover.
- Judge the agent against a short set of real cases that the front desk already handles well.
Why reminders fail when the history is scattered
Most schools track rank in a paper book, a grading spreadsheet, a membership system and the memory of the head instructor. Each source is correct on its own day, and they drift apart over time. A student is promoted in class on a Tuesday, the spreadsheet is updated the following week, and the membership system still shows the old program name. A person at the front desk can ask the instructor before saying anything. An agent cannot, unless the conflict has already been resolved in the record.
The failure is rarely dramatic. A parent receives a reminder to book a test for a belt their child already holds. A student on an approved medical freeze gets a renewal notice. A family grandfathered into an older plan is told a different amount is due. Each message is small, but each one tells the family that the school's records cannot be trusted, and that impression is hard to repair.
What the agent needs to know about each student
Before you connect any tool, write down the fields the agent is allowed to read for every student. A workable set usually includes:
- Current rank, the date it was granted, and the instructor who signed off on it.
- Requirements passed toward the next test, and the date of any scheduled test.
- Membership plan, its billing cycle, and the renewal or end date.
- Approved freezes, holds or exceptions, each with a start date, an end date and the staff member who approved it.
- Message consent, and a guardian contact for every student who is a minor.
- Notes that change what can be said, such as an injury restriction meaning that a test date must be confirmed by the instructor first.
Keep the list short. Every extra field is one more thing that can go stale, and one more thing the agent might repeat to a family.
A step-by-step plan to clean the records
Cleaning the history is slower than building the agent, and it is where most of the reliability comes from. A sensible order looks like this:
- Choose one source of truth for rank and another for membership status, and write down which one wins when they disagree.
- Export the last two years of promotions and check each one against the grading sheets or certificates that were actually signed.
- Standardize program names, rank names and plan names, so the agent does not treat one rank written two ways as two different ranks.
- Mark records with missing dates as missing rather than filling them with a guess.
- Record every freeze and exception with dates and an approver, and close any that have expired.
- Add a consent flag and a guardian contact for each student under age, then check that the flags match what families actually agreed to.
- Assign one staff member as the owner of each record, responsible for corrections and for reviewing the agent's drafts.
Some answers will sit with the head instructor, not the software. Put those questions on a short list and schedule time to answer them, rather than letting the agent make an educated guess in the meantime.
Deciding what the agent sends on its own
The approval boundary is the most important design choice in the project. A useful way to draw it is to sort messages by what happens if they turn out to be wrong.
- Send automatically: reminders about a test date already on the calendar, for students in good standing with no open flags, and renewal windows that match the plan terms in the record.
- Draft for staff: any message that mentions price, late fees, a cancellation, a refund, an injury note or a change to a plan.
- Never send: anything that suggests a student is eligible for promotion before the head instructor has confirmed it.
Illustrative scenario: a children's class has a test scheduled six weeks out. The agent sends the standard notice to families whose children are current and have no flags. For one child whose record shows an open injury note, it creates a draft for the instructor, who confirms whether that child tests on the same date. Nothing about that child's eligibility leaves the school until a person has reviewed it.
The same pattern appears in other businesses built on recurring visits. A studio handling lash and brow rebooking or a clinic running visit recalls faces the same split between routine timing messages and anything that involves clinical or financial judgment. Membership-heavy studios with waitlists, such as yoga studios, also need clear rules about what a message may promise.
Testing the agent before families see it
Before you turn on automatic sending, test the agent on your own history. Build a small set of cases: a student promoted last week, a student on a medical freeze, a family that asked to stop messages, a membership that changed plans mid-cycle, and a renewal that is currently disputed. For each case, write down the right message, or the fact that the right answer is no message at all. Then run the agent on those cases and compare its output with what your front desk would do.
Evaluation practice for AI systems, including the approach described in the OpenAI evals guide, follows the same pattern: fixed test cases, clear pass criteria, and repeated runs after every change. Keep the set small enough that a staff member can review it in one sitting, and add a new case each time a real message goes wrong.
Once the agent is live, log each message with the record it used, the approval it received and the outcome. A monthly review of that log shows whether reminders are landing, where drafts pile up, and which fields keep going stale.
Keeping history intact as the school changes
Schools change hands, instructors leave and software gets replaced. The agent's memory should therefore live in the school's own records, not inside a model's conversation. When the agent needs a fact, it should retrieve the current value from the record, and when a fact changes, the change should be written down with a date and a name. The same principle appears in guidance on context engineering for agents: what the agent sees at each step should be deliberate and limited to what the task needs.
Access should follow least privilege. Each tool the agent uses should carry only the permissions its job requires, and those permissions should be reviewed whenever staff roles change. A reminder agent does not need payment details, and a scheduling tool does not need permission to edit ranks.
Be Human approaches this as a managed service. The first engagement defines one useful role for the agent, the memory and tools it needs, the approval boundaries, and a success criterion the school can observe. Ongoing operations can include monitoring, recurring evaluations, maintenance of connections to your systems, permission reviews and controlled changes when programs or policies change. The work is remote and available in English and Portuguese.
Choosing the first task worth handing to an agent
Start with one role, one record and one kind of message, then widen only after the drafts have been reviewed and the failures understood. Routine, time-bound reminders that need little judgment are usually the right place to begin, and a clean record is what makes them safe.
Take the free Be Human AI diagnostic to find the first task worth handing to an agent in your school.
Frequently asked questions
Can an AI agent send belt-test and renewal reminders without staff review? Yes, for the lowest-risk messages, such as a test date already on the calendar for a student in good standing. Anything that touches price, cancellation, injury or eligibility should go to a staff draft queue first. Write that boundary down so it does not depend on anyone's memory.
What student information should stay out of the agent? Leave out payment card details, full medical histories and anything the family has not agreed to share. Give the agent only the fields a reminder needs, such as rank, test date, plan terms and message consent. Review that list whenever the school changes its software or its policies.
How is this different from the reminder feature in our membership software? Built-in reminders usually run fixed rules on fixed fields, and they often cannot tell that a freeze was approved last month or that a rank was corrected by hand. An agent can read the context around a record and draft a message that fits it, but only if the record is clean and the approval rules are explicit. Many schools use both, with the software sending simple notices and the agent handling the exceptions.
How do we know the agent is working? Measure it against the test cases you built from real history, and review a sample of live messages each month. Useful signs include fewer corrections from families, fewer drafts that staff rewrite heavily, and fewer records flagged as missing or stale. If those signs move the wrong way, pause automatic sending and fix the record before going further.
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.


