Agentic Data & Software Engineering

Knowledge management: when the only person who knows how it works leaves

By Xavier Franco, CEO de Elternativa · Grupo Zemsania
· 5 min read

Written with the help of AI and reviewed by Xavier Franco

A dandelion against the light, with many of its seeds already carried off by the wind and a few still clinging to the stem.
In this article

Almost every company has a system like this. It works, it’s critical, and only one person truly knows how it’s built. They know which process you don’t touch at month-end, why that table has a column nobody uses, and what happens if you change that parameter. While that person is around, all is well. I frame it as one of the three missions of any technology leader: depend less. And depending less starts with knowledge management.

Knowledge lives in heads and conversations

In most teams, what matters isn’t written down. It’s in the heads of a few people, in chat threads, in meetings with no minutes and in documents that stopped being updated.

The 2024 Stack Overflow survey puts it in numbers: 61% of developers spend more than 30 minutes a day searching for answers or solutions to problems. And 30% say that knowledge silos affect their productivity ten or more times a week.

Half an hour a day per person, looking for something someone already knows. Multiply it by your team and by a year.

In practice: ask your team who they call when something fails in your most critical system. If it’s always the same name, you’ve already identified the risk.

The risk isn’t that they leave: it’s that nobody knows

When the key person leaves, through retirement, a job change or simply on holiday, three problems appear:

  • The system freezes. Nobody dares to touch what they don’t understand, and changes get postponed.
  • Everything costs more. Every incident takes twice as long, because first you have to understand and only then fix.
  • Worse decisions get made. Migrating, maintaining or shutting down a system nobody understands is deciding blind.

And there is a less visible effect: that key person is often trapped. They can’t work on anything new because everything goes through them.

AI can make it worse or fix it

Here AI has two faces. Badly used, it multiplies the problem: each person talks to their own AI, what it learns stays in that conversation and the next day it starts from zero. Knowledge doesn’t accumulate; it scatters faster.

Well used, it’s exactly the opposite. If the AI works inside a system that writes down what it learns, every task leaves something behind: a rule, a decision, a mistake not to repeat. And the next task, whoever does it, starts from there.

In practice: check where what your team discovers while working with AI ends up. If it stays in each conversation’s history, it doesn’t belong to your company: it belongs to each person.

What needs to be written down

For knowledge not to leave with people, four things have to live in a system, not in anyone’s memory:

  1. The rules: how things are done in that system, with their version.
  2. The decisions: what was decided, why and what it affects, so it isn’t debated again every month.
  3. The measured errors: what failed, which safeguard watches for it now and whom it alerts.
  4. The documentation: from meetings, from the project and from the code itself, indexed so it can be searched and cited.

And one condition: that this knowledge base is yours. That it lives on your infrastructure and not on a vendor’s, so that knowledge doesn’t change hands if you change vendor.

The key person moves to validating

This isn’t about replacing anyone. Quite the opposite. The person who knows the system adds the most value when what they know gets written down: they review and validate what is documented, correct what isn’t right and give the green light.

Their knowledge stops being a bottleneck and becomes the reference. And they are free to work on what comes next.

In practice: give your key person a few hours a week to validate the documentation of their system, instead of answering the same questions over and over.

When the system is old

If the critical system is also a legacy system, with no documentation or tests, the starting point is different: first you have to extract what it does from the code itself. It is inventoried in read-only mode, and the functional and technical documentation comes from the code, not from anyone’s memory. Then the key person validates it rule by rule.

Where to start

  1. Identify the systems that depend on a single person.
  2. Decide where the knowledge lives, in a database that is yours.
  3. Write down rules, decisions and errors as they appear.
  4. Have the key person validate, instead of answering.
  5. Measure how many queries still reach them after a few months.

How we help

At Agentic Data & Software Engineering we orchestrate the development AI you already have with a method in which knowledge stays in your house: the methodology and the knowledge live in the system, on your infrastructure, so that turnover doesn’t stop the service.

We do it with Agentagile: the project’s rules, decisions, measured errors and documentation are kept in a database you own. The team changes; the method stays. If someone leaves, their rules and what was learned keep working.

Frequently asked questions

What is knowledge management in IT?

It means making sure that what the team knows about its systems (rules, decisions, errors and documentation) is written down, up to date and accessible, and not only in the heads of a few people.

How much time does a team lose searching for information?

According to the 2024 Stack Overflow survey, 61% of developers spend more than 30 minutes a day searching for answers or solutions.

Can AI help with knowledge management?

Yes, if it works inside a system that writes down what it learns. If every conversation starts from scratch, knowledge gets scattered even more.

Where should project knowledge be stored?

In a database owned by the company and hosted on its own infrastructure, so that it doesn’t depend on a person or on a vendor.

Keep reading