Personalized AI infrastructure
Superintelligence for Family Firms: Keep Your Know-How in the Company
Learn how a family business can capture informal know-how in company-owned memory, keep family control of permissions, and pass it to the next generation.

Superintelligence for a family-owned business means giving the company's accumulated know-how a durable home: a memory that AI agents can use, that the family can inspect and change, and that stays under family authority when leadership passes to the next generation. It is not a single product or a forecast about future machines. In practice it starts with the knowledge the founders hold in their heads, moves the company-relevant part into memory the business owns, and attaches permissions so that every agent acts only within decisions the owners have made.
Key takeaways
- Family know-how lives in judgment, exceptions and relationships, and it leaves with people unless it is written into company-owned memory.
- Capture the reasons behind decisions, not only the procedures, because reasons are what a successor cannot rebuild alone.
- Keep the memory, the administrator accounts and the permissions in company names, not in a personal login or a vendor's account.
- Give each agent one role, a limited memory scope and approval boundaries that the owners have signed off.
- Test the memory with the questions a successor will really ask, and review access on a fixed schedule.
Where does family know-how actually sit?
Most of what makes a family business work is not in its manuals. It sits in the reason a supplier is paid early, the customer who gets a phone call before a price change, or the machine that is never started on a Friday because of something that happened years ago. The founder carries this as judgment. The next generation inherits the results without the reasons, and learns the reasons slowly, often by paying for mistakes the founder already avoided.
Some of this knowledge belongs to the company, such as how a product is configured, how a contract renewal is handled or which approvals a certain purchase requires. Some belongs to the family, such as how a disagreement between relatives is handled or what was promised to a long-serving employee. Draw that line on paper before any tool is involved. Company-side knowledge can move into memory the business owns. Family-side knowledge stays with the family and is not handed to an agent by default. For a broader view of succession in this context, see Superintelligence for Family-Owned Businesses: Succession Through Memory.
How do you sort know-how before storing it?
Storing everything is the most common mistake. A memory that holds every email, every conversation and every private remark becomes hard to trust, expensive to maintain and risky to share with an agent. Sort first, then store. Run a short exercise with the owners and the two or three managers who make the most recurring decisions:
- List the decisions made in a normal month, with the information each one depends on.
- Mark each decision as operational, commercial, financial or relational.
- Note who could make the decision today if the owner were unavailable, and what that person would need to ask.
- Mark items that belong to the family privately and exclude them from company memory.
- Name one owner for each remaining item, who will correct it when it goes out of date.
Personal data needs its own rule. If a record describes employees, customers or suppliers as individuals, decide what the company actually needs and keep only that. Brazilian companies can check guidance from the national data protection authority (ANPD) on how personal records should be kept and who may read them. The internal standards for naming, dating and reviewing records are covered in Documentation Standards for a Superintelligence-Ready Company, which is worth applying before any capture begins.
What is a practical plan for capturing it?
A workable capture plan runs in three passes and does not require a large project team. The first pass is interviews. Ask the owner and each key manager to walk through three real decisions from the last quarter, and ask for the reason behind each step. Write the reasons in plain sentences, and note what information the person looked at before deciding.
The second pass is cases. Collect ten to fifteen real situations, including the awkward ones, and write down what the business did and why. These cases become the test set later, so keep them specific. The third pass is review. Give the notes to someone who was not in the interviews and ask them to make a sample of the decisions using only the written material. Every place where they would have to call the owner marks a gap to fill.
Illustrative scenario (a hypothetical case, not a real company): a third-generation distributor of packaging materials hands purchasing to a daughter who has worked in sales. During the interviews, the founder explains that one supplier receives early payment in winter because its warehouse closes for two weeks each January, and that a short delay there once stopped three client orders. None of this appears in the purchasing procedure. Once the reason, the trigger and the owner are written down, the new purchaser can make the same call, and the agent can flag the same risk before an order is placed.
How should family control work inside the memory?
Control has three parts: who owns the memory, who may change it, and what an agent may do with it. Ownership means the company holds the storage, the accounts and the administrator roles under company names. Those roles should be recorded in writing, with the people who hold them, and reviewed when responsibilities change. The OWASP authorization guidance describes the underlying principle clearly: deny access by default and grant only what a role needs. Apply that to agents. An agent that drafts supplier reminders does not need access to payroll notes, and an agent that reads contracts does not need permission to send them.
Write the approval boundaries as three short lists: actions the agent may take alone, actions it may only propose for approval, and decisions only an owner may make. Keep the first boundaries tight. Widen them only after the team has seen the agent handle real cases correctly over time. Access should be reviewed on a schedule, and the steps for that review are set out in Quarterly AI Agent Permission Review: Steps and Evidence.
How do you know the memory still works?
A memory is only useful if it gives correct answers when the successor needs them. Test it with the cases from the second pass. Each time the agent answers, record which memory items it used. That trail lets you trace a wrong answer to a stale note instead of blaming the model or guessing. The OpenTelemetry documentation on observability concepts explains why this kind of trail matters for software systems in general, and the same logic applies to agent work.
Most business agents do not learn a company by being retrained. They retrieve approved records when a question arrives, the pattern described in AWS's explainer on retrieval-augmented generation. This makes the memory easier to correct: when a record changes, the next answer changes with it, and the change is visible in the log.
Re-run the same case set on a recurring schedule, usually quarterly for a stable business. When a case fails, decide whether the memory needs correcting, the permission needs adjusting, or the question sat outside the agent's role. Keep a short log of these decisions. Successors inherit the log along with the memory, and it shows them why the company reasons the way it does.
How does a managed approach keep this under family control?
Be Human designs, implements and operates personalized AI infrastructure for companies, with agents that keep their identity, memory, channels, tools, permissions and operating context under human control. A first engagement defines one useful role, the memory and tools that role needs, its approval boundaries and an observable success criterion. Managed operations continue that structure: monitoring watches how the agent behaves, recurring evaluations re-run the case set, permission reviews check who and what has access, and changes go through a written request and approval step. The NIST AI Risk Management Framework treats this kind of ongoing governance as a continuing activity rather than a one-time setup, which fits the way family memory has to be maintained. The practical result for a family business is that the memory, the accounts and the decision rules stay documented and owned by the company, so a change of leadership does not depend on one person's recollection of how the system was built.
If you want to find the first task in your company worth handing to an agent, take the free Be Human AI diagnostic.
Frequently asked questions
Does superintelligence matter for a family business that is not a technology company? It matters wherever a business depends on accumulated judgment that is not written down. A family manufacturer, distributor or service firm usually holds more of that judgment than its documents show. The practical question is not the technology category but whether the company can keep its reasoning when people change roles.
Should the family put personal or private matters into company AI memory? No. Private family matters and personal details about individuals should be kept out of company memory unless the company has a clear, lawful need for them. Keep the memory limited to decisions and records the business needs to operate. Review personal data rules with a qualified adviser, and check the guidance of the relevant data protection authority.
Who should own the AI memory when the next generation takes over? The company should own it, meaning the storage, the accounts and the administrator roles sit under company names rather than a personal login or a vendor account. The family decides who holds each role, and those roles are written down and reviewed during the handover. This keeps control with the owners whoever is leading the business at a given time.
How long does it take to know whether the memory works? It depends on how many decisions you capture and how often the team uses the agent. The useful signal is whether a person who was not part of the interviews can make the sampled decisions from the written material alone, and whether the recurring case set keeps passing as the business changes. Judge the memory by those tests, not by how much has been stored.
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.


