Superintelligence in Basic Education: What Schools Can Delegate to AI

Superintelligence in Basic Education: What Schools Can Delegate to AI

Which school and admissions tasks can be delegated to AI, which must stay with teachers and families, and how to set limits, records and human review.

•

Want to apply this in your company?

Book a call with Juan →

Superintelligence, in the sense the term carries in current debates, is not a system a school hires to run on its own. What can already change the daily routine of basic education schools are narrow AI systems: answers to family questions, triage of administrative requests, drafts of school notices and support for planning. The practical rule is to delegate to AI what is repetitive, verifiable and easy to undo, and to keep with teachers, the pedagogical team and families everything that involves assessment, care for the child and decisions about a student's school path.

Key takeaways

  • AI should handle repetitive administrative and communication tasks, with answers logged and reviewed whenever there is risk.
  • Assessment, student reports, conduct, referrals and retention decisions stay under the responsibility of teachers and the pedagogical team.
  • Data about children and adolescents requires extra care, based on Brazil's LGPD (the general data protection law) and, in many cases, specific consent from at least one parent or guardian.
  • Before using any tool, the school needs to map the process, define what the AI must not do and name an internal owner.
  • A small pilot with one agreed metric and a decision log is usually more useful than a broad rollout at once.

What changes in the school when AI enters the routine

In practice, the most visible change is not in the classroom but behind it. The administrative office receives the same kinds of questions many times a week: which documents to bring for enrollment, what the meeting calendar looks like, how to request an attendance certificate, whether a certain shift has openings. These questions take up service time and, when answered late, create friction between families and the school.

A well-configured AI agent can answer these questions using documents the school has approved, pass anything else to a person and keep a record of what was said. The real gain is freeing the team for work that requires judgment. This does not make the school more impersonal. It means a conversation with a family about a learning problem no longer competes with a queue of questions about schedules.

The same logic applies to communication. Notices, reminders and summaries can be drafted with AI support, as long as the administrative office or the coordination reviews the text before it is sent. The most common mistake is treating the draft as a final answer, especially when the text mentions dates, fees, assessment rules or individual situations.

What can be delegated to AI in administration and communication

The tasks best suited for delegation are those with a verifiable answer in an official document and a low or easily corrected cost of error. In general, these fall into that category:

  • Answering frequently asked questions based on the calendar, required documents and office hours published by the school.
  • Running the first triage of requests (certificates, second copies, transcripts) and routing each one to the person responsible.
  • Sending meeting reminders, document deadlines and enrollment deadlines, when the family has already given permission for this kind of contact.
  • Organizing meeting schedules with teachers or coordinators, within the windows the team has defined.
  • Preparing draft notices that the administrative office reviews before publishing.

What these tasks share is a clear boundary. The agent does not decide whether a student can enroll after the deadline, does not grant exceptions and does not interpret rules the school has not formalized. When a question falls outside that scope, the correct answer is a handoff, not a well-written guess. For an example of how this logic works in enrollment and parent communication, see AI for Schools: Admissions and Parent Communication Guide.

What should stay with teachers, coordination and families

Some decisions should not be delegated, even when a tool can produce a convincing text. Learning assessment, descriptive reports, interpreting learning difficulties, referrals for pedagogical or psychological support, disciplinary measures, retention decisions, adaptations for students with disabilities and sensitive conversations about family or health belong to people who know the student, carry institutional responsibility and can be questioned.

The reason is not only bureaucratic. A teacher knows that a student has gone quiet after a change at home, notices when a low grade comes from a hard week and can listen to the family. A text-processing system does not have that context, and its answer can sound certain even when it is wrong. Decisions that affect school life also need an identifiable person accountable for them, and an AI system is not one.

The LGPD also covers the review of decisions made solely on the basis of automated processing that affect the interests of the data subject. For that reason, any workflow in which a system classifies, scores or refuses a request needs a path for a person to review the decision. This point is covered in LGPD Automated Decision-Making: The Right to Review (Art. 20).

Families also have a role that should not be replaced. They need to know when they are talking to an agent, what information the system collects, how to request human support and where to complain. A school that hides its use of AI loses trust faster than one that explains it clearly.

Student data: the care a school cannot outsource

Student data is, to a large extent, data about children and adolescents. The LGPD sets specific rules for this processing, including the requirement that it serve the best interest of the data subject and, in many cases, rely on specific and highlighted consent from at least one parent or legal guardian. Other legal bases may apply, so the school should confirm with its legal advisor or data protection officer which basis fits each workflow, rather than assuming a generic consent form solves everything.

In practice, three safeguards matter. The first is minimization: the agent should access only what the answer requires, not the full student file. The second is prior approval of tools: before any data leaves the school for an external service, the school needs to know where it is processed and how long it is stored. If the data is sent to another country, the analysis must consider transfer rules, a topic covered in International Data Transfer Under LGPD When Using AI Providers. The third is records: every relevant conversation should be retrievable by a responsible person.

For official and current guidance on data protection, the reference is the Brazilian National Data Protection Authority (ANPD), available at gov.br. For managing risks in AI systems, the NIST framework offers a useful structure for schools and networks that want to organize the topic without relying on technical jargon.

A decision guide for each task

Before delegating any task, the team can run through five simple questions. They do not replace legal analysis, but they prevent most design mistakes:

  • Does the task repeat often and follow written rules? If not, it probably needs to stay with a person.
  • Can the correct answer be checked against an official school document? If there is no document, the AI should not answer with authority.
  • Is an error easy to correct without harm to the student or family? If an error would affect school life, human review is mandatory.
  • Does the task involve judgment about performance, behavior, health or family situation? If so, it stays with teachers or the pedagogical team.
  • Does the task require sensitive data or data about children? If so, the legal basis, the consent and the tool must be validated before any use.

A task that passes all five questions can be delegated with supervision. A task that fails any one of them should stay with a person, even if AI helps prepare supporting material.

Illustrative scenario: a school office with a questions agent

Illustrative scenario, not based on any real school: a private elementary school wants to reduce the message queue at its administrative office during enrollment season. The team decides to start with one workflow, questions about documents and the calendar, and keeps out of scope any request for exceptions, scholarships, transfers or complaints.

The plan follows a simple sequence:

  • List the most frequent questions from recent conversations, without identifiable data, and separate those with documented answers from those that depend on a decision.
  • Validate with the principal which official texts the agent may use as sources, and remove any outdated information.
  • Define what the agent does not do: it does not grant extra deadlines, does not interpret individual cases and does not handle health or conduct topics.
  • Set up handoff to a named person in the office during fixed hours, with a notice to the family about when to expect a reply.
  • Log every handed-off conversation and review a sample each week with the coordination.
  • At the end of the test period, compare the agreed metric, for example the time until the first correct answer, and decide whether to expand, adjust or stop the workflow.

What this scenario shows is not the efficiency of the tool but the discipline of the process. The school decides in advance what the AI may answer, what it must hand off and who is accountable for the result.

How to apply the method in a school or school network

In our process design work with AI, the path that tends to hold up has four steps: map, limit, pilot and hand over. Mapping means describing the current process, with volume, time and friction points, before discussing any tool. Limiting means defining which data and which actions the system can access, and what it must never do. Piloting means testing one real workflow, with logged conversations and a success metric agreed before the start. Handing over means giving the workflow to the team with instructions, owners and a criterion for deciding whether it will be expanded, adjusted or switched off.

This method does not eliminate risk, but it makes risk visible. A school that follows this sequence knows when an error happened, who was alerted and what changed afterward. For teams that need to decide without technical jargon, the Superintelligence Glossary for Executives, Without the Jargon helps align terms before budget and accountability conversations.

When the school uses WhatsApp to reach families, it also needs care with messages started by the institution. The topic of permission and message templates is covered in WhatsApp Outbound Messages: Templates, Opt-In and Best Practices.

For a broader reference on principles, the OECD maintains its AI Principles, which emphasize transparency, accountability and respect for human rights. They do not say which tool to buy, but they are useful for checking whether the chosen design makes sense for the school community.

Frequently asked questions

Can AI replace teachers in basic education? It is not advisable. AI can support preparing materials, organizing routines and communication, but assessment, student follow-up and pedagogical decisions require knowledge of context and the accountability of an identifiable person.

Does the school need parental consent to use AI with student data? It depends on the legal basis that applies to each workflow. Under the LGPD, processing data about children and adolescents has specific rules, and in many cases specific and highlighted consent from at least one parent or guardian is required. The school should confirm this with its legal advisor or data protection officer before activating any tool.

Where should a small school office start? The best start is mapping the most frequent questions and separating those with documented answers from those that depend on a decision. Then choose one workflow, define what the system will not do and agree on a simple metric, such as time until the first correct answer, to decide later whether to expand.

How do you know whether the AI's answer is correct? The school should define the official sources the agent may use and review a sample of conversations on a regular schedule. Answers that cite deadlines, fees, assessment rules or individual situations should always pass through a person, and any error found should lead to an adjustment in the agent's reference material.

Sources