
Superintelligence for Notary Offices: AI Triage and Deadline Reminders
How an AI agent can triage requests and track deadlines in a notary office, and why legal decisions and final review stay with the notary and staff.
Want to apply this in your company?
Book a call with Juan →Superintelligence, in the sense used in research, is not a tool a notary office switches on. What a notary office can deploy today is a narrow AI agent that triages requests, checks whether the documentation is complete, organizes the queue, and reminds the team of deadlines and open items, routing each case to the staff. The qualification of the deed, the interpretation of the law, and the final act remain with the notary and the clerks, who answer for the delegated public function set out in the Constitution for notarial and registry services. This article explains what an agent can do safely, where it must stop, and how to start with a single flow.
Key takeaways
- Superintelligence is a research concept. In practice, a notary office runs narrow agents with defined tasks, inputs and outputs.
- The agent can sort requests by type, document completeness and urgency, without giving an opinion on whether the act is valid.
- Deadline and pending-item reminders are the most predictable gain, as long as people define the calendar rules.
- Legal qualification, requirements and final acts stay with the notary and staff, and every agent action must be logged.
- Start with one type of request, with minimal data, logs and an agreed success criterion before expanding.
What the word superintelligence means for a notary office
The term appears in headlines and in debates about the future of technology, but a management team needs a more useful question: what does a system do with a defined input, and who answers for the result? For that question, it helps to separate speculation from operation. The distinction between general artificial intelligence and superintelligence helps move the topic out of science fiction, and the difference between a language model and an agent that acts inside systems also deserves attention. If you want to go deeper into that separation, read Difference Between AGI, ASI and AI Agents: What You Can Run.
In the daily routine of a notary office, the useful discussion is different: which repetitive tasks consume the team, which data comes in, which responses go out, and where responsibility changes hands. A well-designed agent is not glamorous. It reads the request, compares it with a list of required documents, asks what is missing, records what it did, and hands over an organized file to the person who decides.
What an agent can triage safely
Triage is the first job worth delegating because it is classification and presence checking, not legal analysis. A typical flow can separate requests by type of act, verify that mandatory fields are filled in, check whether attached documents are legible, confirm that the requester provided the data the office itself requires, and sort the queue by deadline.
A triage checklist can include the following items, always based on rules the team already uses:
- Request type identified and linked to a document script.
- Requester identification present and legible, according to the office's rule.
- Required documents listed and attached, with any gap stated objectively.
- Deadline recorded, if the office itself set one for the request.
- Request flagged for human review when the stated data and the documents do not match.
What stays off this list matters as much as what goes on it. The agent does not decide whether a deed is valid, does not say whether an act is permitted, does not promise an outcome, and does not suggest legal strategy. It organizes and flags. That boundary must be written into the flow design before any testing begins.
How deadline reminders work without becoming decisions
Reminding is different from deciding. The notary office sets the rules: which deadlines apply, when a requirement expires, what message should be sent, and who is alerted when a deadline approaches. The agent executes those rules, sends reminders within what has been approved, and logs each dispatch. When there is no reply, it does not invent a solution. It escalates to the responsible person.
Illustrative scenario: a requester submits a request through WhatsApp and does not attach a supplementary document. The agent identifies the gap using the checklist, sends the standard message approved by the team, logs the send, and schedules a reminder. If there is no reply by the deadline the office has set, the file appears for a clerk with the full history. No step involves judging whether the document would be sufficient for the act, because that assessment belongs to the team.
On the messaging channel, timing rules matter. Conversations outside the WhatsApp customer service window require approved templates and, in many cases, prior consent from the recipient. Before designing any automatic reminder, review WhatsApp 24 Hour Customer Service Window: How an AI Agent Should Act and WhatsApp Outbound Messages: Templates, Opt-In and Best Practices.
Where the legal decision stays with the notary
The delegated notarial and registry functions are exercised by the notary and the authorized staff. That does not change because a tool is present. The agent can be the first to read a request, but it cannot be the last word on any act. For that reason, the flow design needs clear triggers for handing cases to people.
Signals that should always send a case to the team:
- A mismatch between the name, document or data provided and what appears in the attachments.
- Doubt about the requester's capacity, representation or identity.
- A request that does not fit any script on the checklist.
- A request for an exception, urgency with a claimed loss, or a reference to a court decision.
- A complaint, dissatisfaction, or any sensitive content.
- Any question that asks for interpretation of the law or guidance on the outcome of an act.
In these situations, the agent should state clearly that the case has been passed on, without attempting to answer the substance. The message to the requester should name the next step and the team's contact channel.
Personal data, LGPD and recording actions
Notary files contain personal data and often sensitive data. Brazil's LGPD requires a clear purpose, necessity and security in processing, and the National Data Protection Authority publishes the main guidance on the subject. In practice, the agent should read only what triage requires, each request file should have restricted access, and the office must know where the data is processed.
If the model or the hosting tool sits outside Brazil, international transfer enters the analysis. See International Data Transfer Under LGPD When Using AI Providers. When triage meaningfully affects the progress of a request for the person involved, it is also worth reading LGPD Automated Decision-Making: The Right to Review (Art. 20), because the right to request human review needs a workable operational path.
As for records, a useful log answers four questions: what the agent read, what it suggested or classified, which messages it sent, and who made the final decision. Without those answers, the office cannot audit or correct the flow.
How to start: a pilot with one type of request
Juan Carlo works with a structured method: map the current process and its baseline, limit data and actions, pilot a real flow with logs and an agreed metric, then hand over to the team and decide whether to expand, adjust or stop. Applied to a notary office, the steps look like this:
- Step 1: map the current flow, from receiving the request to the final act, including deadlines and the points where requests stall today.
- Step 2: list the types of request and turn the office's official checklist into written rules, validated by the notary.
- Step 3: limit the agent's access. It should read the queue and the necessary attachments, with no permission to change records or issue acts.
- Step 4: define a success metric before the pilot, such as time to first response, the share of requests complete on first submission, or the number of missed deadlines, and measure the baseline in the current process.
- Step 5: test the flow in a controlled environment, with simulated cases that include incomplete documents, mismatched data and off-pattern messages. Testing with mocks and a sandbox helps at this stage, as shown in Mock Tools for Testing AI Agents: Safe Sandbox Testing Guide.
- Step 6: run the pilot on a single type of request, with daily log review by the team, and at the end decide: expand, adjust or stop.
The rule is to start small. An office that tries to automate every act at once loses the ability to understand what changed.
Risks that need control from the start
The risks of an agent in a notary office are well known and can be mitigated. The first is instruction injection in documents or messages: an attachment or a text sent by the requester may contain commands trying to change the agent's behavior. The OWASP project maintains a list of risks for applications built on language models, and the practical response is to treat all received content as data, never as instructions, and to give the agent no actions beyond triage.
The second is incorrect extraction, when the agent reads a number or a date wrongly. The fix is to require human verification on any field that affects the act and to log the discrepancy. The third is the invented deadline: the agent should never calculate a legal deadline on its own. It uses only the rule registered by the office. For general governance, NIST publishes an AI risk management framework that the team can use as a reference to document its decisions.
When an office wants a model hosted locally for data reasons, the infrastructure choice changes the size of the risk but does not remove the need for rules and logs. For that path, read Running Ollama Local Models in a Company: A Practical Guide.
Frequently asked questions
Can an AI agent give legal advice at a notary office? No. The agent can organize requests, check whether documents listed by the office were attached, and remind the team of open items. Legal interpretation, qualification of the deed and the decision on the act belong to the notary and the authorized staff, and the flow should route those cases to them.
Is superintelligence something a notary office can buy off the shelf? There is no superintelligence product for this use. What gets deployed are narrow agents with defined tasks, written rules and human review. Thinking in these terms avoids empty promises and makes clear what will be measured.
Which data should the agent access to triage requests? Only what triage needs, such as the request type, mandatory fields, and the attachments relevant to the checklist. Access should be read-only, with every query logged, and the office should document the purpose of processing under the LGPD, including where the data is processed.
Where should a notary office begin a project like this? Start with one type of request that has high volume and a well-defined checklist, map the current flow, and agree on a success metric before any automation. A pilot with logs, daily review and a clear decision point at the end will show whether expansion makes sense or whether the process needs adjustments first.