When Someone Leaves, Their Knowledge Leaves Too: Internal AI Clones as Company Memory

When a key person leaves, the documents stay and the reasons go. The procedure is still on the shared drive. What is lost is why it is done that way, which exceptions were already settled, and which traps the person learned to avoid. An internal AI clone keeps that context: one clone per person, owned by the company, that records decisions with their reasons as the work happens, not in an exit interview.

We did not start from theory. Our founder, Nabil BA-MOH, built Klyr, an AI clone, for himself, and Asio runs on it every day. What follows is what we actually record in it, and why.

What exactly leaves when a key person leaves?

Rarely the procedure. Usually the judgement around it.

Picture a finance controller who leaves. She takes with her the knowledge that one entity always books a certain flow late, that a past exception was agreed with the auditors, and that a report looks wrong every March for a known reason. None of that is in the procedure. It lives in her head, in old emails and in the habit of the people around her.

Documentation keeps the what. The why leaves with the person. And the why is what the next person needs, because it is what tells them whether a rule still applies.

Why doesn’t documentation solve it?

Because documentation is written after the fact, by someone who already knows the answer, for a reader they imagine. It records the rule and drops the reasoning, because the reasoning felt obvious to the writer.

It is also written once. The person keeps learning after the page is written, and the page does not. Six months later the page describes a process that has drifted, and nobody knows which half is still true.

The capture has to happen at the moment the knowledge is created: when the decision is made, when the trap is hit, when the exception is settled. That is the moment people have the least time to write anything down, which is exactly where an AI clone helps.

What does an internal AI clone actually keep?

What you give it, with the reason attached. In our daily use, four kinds of things go into it:

• Decisions, with their date and their reason. When a question is settled, the clone records the answer, who decided, when and why, and marks it as settled so it is not asked again. Re-opening a settled question costs more than people think: it costs the time, and it costs trust.
• Traps, written down the day they are hit. When we drew a wrong conclusion, we record the wrong conclusion itself, not just the right answer. Our clearest case is an agent that searched our ERP data on the wrong field and reported “nothing found”. That trap, and the method we use to verify what an AI agent reads, now sit in the routine that would have hit it again.
• Facts that expire, marked as such. Some facts hold for a year. Others change every month: a list of people, a code that rolls over, a date that shifts. We mark the second kind “check each time”, and record the date each fact was last verified.
• Routines. Work that repeats, such as our monthly time sheet checks, lives as a routine with its steps and its checks. It does not depend on one person remembering the sequence.

The clone does not decide what is true. It keeps what a person decided, and says when it was decided.

How do you keep a company memory from filling up with wrong answers?

By separating what has been decided from what has only been written down.

Our rule is strict: a note is a draft until a person says otherwise. An unreviewed note can be used to think. It is never presented as a fact, and it never travels into a document, a message or a client-facing sentence without someone confirming it first.

The second protection is control. On Klyr, you can read, edit or wipe any memory at any time. A memory that nobody can inspect is not company knowledge, it is a liability.

Who owns the clone when someone leaves?

The company. On Klyr for Teams, every person in the company gets their own clone, and the company owns and controls it. What people build stays with the company, even when they leave.

That ownership is what turns a personal tool into company memory. A personal AI subscription bought by an employee leaves with the employee, with everything it learned. A clone the company owns stays. It is the same logic as a mailbox or a shared drive: the account belongs to the company, the work done in it belongs to the company.

Two limits are worth stating plainly. A clone keeps only what it was given, so a person who never used it leaves nothing behind. And it keeps knowledge, not authority: the next person still has to decide whether a past decision still holds.

Where does this fit in an AI plan for a group?

Next to the ERP, not inside it. The ERP holds the transactions. The clone holds the context around them: why an entity is set up a certain way, which arbitration was made at the last close, which workaround exists and why.

In a group, that context is spread across more people and more entities, so it is lost faster. That is why internal clones are one of the pieces in how we approach AI for multi-entity groups, alongside agents that read and check ERP data.

The Asio Services way

We run our own company on a clone before we talk about it, and we apply the same discipline we would ask of a client. Decisions are recorded with their reasons on the day they are made. Traps go in the day they are hit. Facts that expire carry the date they were checked. And nothing unreviewed is ever passed off as fact.

If you want to know what your company would lose if two key people left next quarter, and how to keep it, start with a clarity session.

FAQ

What is an internal AI clone?

An AI that keeps the context of one person’s work: their decisions and the reasons behind them, the traps they have hit, the routines they run. In a company setting, each person has their own clone and the company owns it.

Does an AI clone replace documentation?

No. Documentation records the rule. A clone records the reasoning and the exceptions as they happen, which is the part documentation usually loses. The two work together.

What happens to the clone when an employee leaves?

On Klyr for Teams, the company owns and controls each clone, so what the person built stays with the company. The clone keeps what it was given; it does not invent what the person never recorded.

Can a company memory spread wrong information?

It can if drafts are treated as facts. Our rule is that a note stays a draft until a person confirms it, and that every memory can be read, edited or wiped at any time.