
AI for Robotics Schools: Classes, Kits and Logged Exceptions
AI for robotics schools:ur an agent that handles enrollment questions, class slots and kit restocking, and logs every exception for staff to decide.
Want to apply this in your company?
Book a call with Juan →For a robotics school, the safest way to handle enrollment, class scheduling and kit restocking is an AI agent that resolves routine cases on its own, within rules the school has written down, and sends every exception to a person while keeping a record of the decision. The agent handles first contact, checks seat availability and tracks kit inventory, while the coordination team approves anything outside the agreed terms. The real gain is less about speed and more about the record, which shows who decided what, under which rule, and why.
Key takeaways
- Separate fixed rules (age range, prerequisites, class capacity) from cases that need judgment (schedule changes with a fee difference, discounts, transfers).
- Let the agent check availability and log requests, without promising a seat before confirmation.
- Treat kit restocking as inventory control with a reorder point, not as a guess based on how many students are in the room.
- Every exception creates a record with the owner, the reason, the data used and the reply sent to the family.
- Start with one program or one unit, agree on a success metric before launch, and only then decide whether to expand.
Why robotics enrollment gets messy faster than it looks
A robotics school receives requests that do not look like those of a regular school. Classes have limited capacity, levels may depend on a previous assessment, physical kits have small parts that get lost or broken, and families ask to change time slots almost every week. When that information is spread across a spreadsheet, a WhatsApp group and the front desk's memory, every request becomes a new conversation.
When people talk about superintelligence in this context, what works today is an agent with clear limits. It performs repetitive tasks with verified data and hands back to a person whatever is not its call to make. That difference is explained in Superintelligence in Practice for Companies: What an Agent Does Today.
The risks are specific too. An agent that promises a seat that does not exist creates frustration and rework. One that grants a discount without authorization sets a precedent. And because most conversations involve information about children and their guardians, any automation should collect and use only what is necessary, as Brazil's data protection law (LGPD) requires. For how the law applies to automated decisions, see LGPD Automated Decision-Making: The Right to Review (Art. 20).
Map the process before configuring any agent
Before writing a single instruction for the agent, find out how the school actually operates. A simple four-step plan is usually enough:
- List the types of request received in recent weeks, including the confusing cases, not only the ideal flow from the handbook.
- For each type, note who decides today, in which system or file the decision is recorded, and how long the reply takes.
- Mark what is a rule (can be automated) and what is an exception (needs a person).
- Check where the data lives: class spreadsheet, school management system, WhatsApp inbox or the kit control notebook.
This map is the foundation of the method we apply: map the process and the baseline, limit the data and actions, pilot a real flow with records and an agreed metric, then hand the operation over and decide whether to expand, adjust or stop. Without the map, any agent will invent rules the school does not have.
Classes: the agent checks and proposes, but does not confirm exceptions
For class management, the agent can safely do three things: show the classes that match the child's age and level, display current occupancy, and hold a seat for a short period until the family confirms. Final confirmation, and any change that affects billing, should go through the coordinator or through a system rule the school has approved in advance.
Illustrative scenario: a family asks to move their child from a Friday class to a Tuesday class. The agent confirms that the Tuesday class has space and that the child's level fits. Because the change affects the monthly fee, it does not confirm. It opens a record with the request, the verified data and the proposed change, and sends it to the coordinator. The coordinator approves, declines or proposes another slot, and the decision is noted with its reason.
A simple decision guide helps the team configure this behavior:
- Request within the rules and with space available: the agent holds the seat, sends the agreed confirmation and logs it.
- Request within the rules but no space: the agent offers the waiting list, records the acceptance and notifies the family when a seat opens.
- Request outside the rules (age, prerequisite, expired deadline): the agent explains the published criterion and passes the case to the coordinator.
- Request with financial impact (discount, swap with a fee difference, refund): always a human decision.
- Complaint or sensitive matter: immediate handover to a person, with a summary of the conversation.
WhatsApp has its own rules for messages sent outside the customer service window. Before scheduling automatic notices about an opening, read WhatsApp 24 Hour Customer Service Window: How an AI Agent Should Act.
Kit restocking: inventory with a reorder point, not guesswork
Robotics kits are expensive to replace and easy to deplete. Each part has a different use, and an advanced class consumes more components than a beginner class. The agent can track this with simple rules: a minimum stock level per item, a count recorded at the end of each class, and an alert when any item falls below its minimum.
What happens after the alert matters most. The agent can assemble the restocking list, group items by known supplier and attach the consumption history. The purchase, however, should be approved by a person. An agent that places orders on its own can repeat a counting error many times before anyone notices.
Every damaged kit or missing part should also generate a record, with the class, the date and the responsibility as the school defines it. This protects the operation and keeps replacement charges from turning into a dispute without a basis.
What to record for every exception decision
The record is what separates an operational agent from an assistant that answers unpredictably. Each exception should store, at minimum, the following fields:
- Date, time of the request and the channel it came from.
- The original request text, with only the strictly necessary data.
- The rule applied and the version of that rule in force at the time.
- Who decided, and whether the decision was made by the agent within a limit or by a person.
- The reason for the decision, in one sentence.
- The reply sent to the family and the deadline for a response.
This record also answers the question that comes up after any problem: what happened, and who decided. When a guardian asks for a review of a decision supported by automation, the LGPD provides that right. For managing risks in AI systems, the NIST AI Risk Management Framework is a useful reference, and the OWASP guidance on authorization is a good starting point for designing the agent's permissions.
Limit what the agent can reach from the start. It should read class data and write only to request records, never to the financial registry. Access is much easier to grant later than to take back after a mistake.
How to pilot in one program and decide the next step
The pilot should be small and real. Choose one program, for example the beginner class, and an operating period long enough for real exceptions to appear. Before the agent contacts any family, test the rules with simulated cases in a sandbox. The process for testing with mocks and sandboxes is described in Mock Tools for Testing AI Agents: Safe Sandbox Testing Guide.
The success metric should be agreed before the start. Good choices are the time to first reply, the share of requests resolved without rework, and the number of exceptions that reached the coordinator with a complete record. If the metric does not improve, the problem is usually in the process map, not in the technology.
At the end of the pilot, the school chooses among three paths: expand to other programs, adjust the rules and repeat the pilot, or stop. Stopping is a legitimate outcome too. The goal is an operation the team understands and can maintain, not an agent that appears everywhere.
Frequently asked questions
Can the agent confirm enrollment on its own?
It can confirm requests that fall within the published rules, with a proven open seat and a recorded reservation. Payments, discounts, age exceptions and prerequisite exceptions should go to a person. This split reduces errors without slowing down the first contact.
How does the agent know that a request is an exception?
It compares the request with the rules the school has published, such as age range, level, capacity and deadlines. When a data point does not match, or when the request involves money or a complaint, the case goes to the coordinator's queue with a summary. The rules must be written down and kept up to date, because the agent can only follow what is documented.
Does the school need a school management system to start?
Not always. Many schools start with a well-organized spreadsheet and a clear contact channel, as long as the source of the data is clear and someone is responsible for keeping it current. A management system becomes important when the number of classes and requests grows and the spreadsheet ends up with more than one version of the truth.
How should children's data be handled during service?
Collect and record only what the request needs, such as the student's name, the class of interest and the guardian's contact. Define who can access each record and how long it is kept. Before going live, review the guidance published by the ANPD, and if there is a legal question, get advice from a qualified professional.