Beyond Knowledge Management: Managing Context

For decades, we have worked to improve how organizations manage knowledge. We have established knowledge bases, document repositories, wikis, intranets, and learning platforms. We document processes, decisions, meeting notes, and lessons learned. Yet many organizations continue to experience the same problem over and over again. When key individuals leave, much of the understanding leaves with them. Not necessarily the knowledge.

The context.

It was only when I lost access to an AI assistant I had collaborated closely with over a long period of time that I truly understood how important this distinction is.

Knowledge and Context Are Not the Same

Let us start with a simple observation.

Knowledge is relatively easy to store. Documents, presentations, source code, architecture diagrams, decisions, and articles can be archived, searched, and retrieved long after they were created. Most organizations have invested significant resources in building systems designed to do exactly that.

Context is something different.

Context is not only about what was decided, but why the decision was made. It includes which alternatives were considered, which risks were discussed, which compromises were accepted, and how the people involved understood the situation at the time the decision was made. Two people can read the same decision document and arrive at completely different conclusions if the surrounding context is missing.

This is not a new problem. I have previously written and reflected extensively on how organizations lose important decision history over time. The decision itself often remains.

The reasoning disappears.

When the AI Suddenly Disappeared

During a transition between different user accounts, I encountered a new variation of the same problem.

I did not lose any code. I did not lose any documentation. I did not lose any articles. Everything was safely stored in GitHub and Knowledge Hub. Yet it still felt as though something important had disappeared.

Over months of collaboration, the AI assistant had gradually built a substantial understanding of the project, the architecture, the ways of working, personal preferences, historical context, and the many discussions that had shaped the direction along the way. Much of this had never been documented explicitly. It existed only within the dialogue.

When the dialogue disappeared, a large part of the context disappeared with it.

The experience led me to ask a question I probably should have asked much earlier:

How do we ensure that context survives when the AI, the chat, the account, or the tool is replaced?

Organizations Have Had the Same Problem All Along

The interesting thing is that this is not really about AI.

AI simply makes the problem visible.

Many organizations believe they are in control because they have documented the decision. What is often missing, however, is the history behind that decision. The reasoning. The alternatives that were considered. The discussions that took place. The assumptions that formed the foundation for the conclusion.

The result is that organizations are forced to learn the same lessons repeatedly. New employees ask the same questions. Old challenges return. Previous discussions need to be repeated. History starts repeating itself.

Not because people are incapable.

But because the context disappeared.

From Knowledge Management to Context Management

Traditionally, we have talked about Knowledge Management. The goal has been to make knowledge available, structured, and searchable.

That is still important.

But I am beginning to believe that it is no longer sufficient on its own.

Increasingly, we collaborate with AI systems that are gradually introduced into projects, build understanding over time, participate in analysis, and contribute to decision-making processes. In that environment, it is no longer enough to ask how we store knowledge.

We must also ask how we manage context.

Because the challenge is not only preserving what we know.

The challenge is preserving why we know it.

The Knowledge Hub Case

This insight led to an interesting evolution of Hugo Knowledge Hub.

Originally, Knowledge Hub was intended to be a personal knowledge repository. A place for articles, presentations, frameworks, decisions, and lessons learned. Over time it became clear that this only solved half the problem.

If the goal was to make it possible to rebuild understanding over time, the platform also had to function as a persistent context repository.

The result was the creation of a dedicated AI Context area.

Not because AI needs more documents.

But because AI needs understanding.

This area documents project context, working principles, design decisions, technical conventions, current priorities, and changes that are not yet visible elsewhere in the documentation. Architectural decisions are documented using a consistent format that describes the problem being solved, which alternatives were considered, why they were rejected, and what consequences the decision is expected to have.

Not because such a format can capture everything.

But because it captures most of it and makes the rest visible as a conscious gap rather than an invisible hole.

The goal is simple.

If a new AI assistant joins the project tomorrow, it should be able to reconstruct a large portion of the context without having to recreate months of conversations.

The Most Important Lesson

The biggest mistake we can make is trying to preserve every single chat.

That does not scale.

And it does not solve the real problem.

Instead, the goal should be to identify which context actually matters. Which decisions have been made. Which principles apply. Which ways of working are effective. Which assumptions form the foundation of current thinking.

When these things are documented explicitly, both people and organizations become less vulnerable.

The same applies to collaboration with AI.

Context Must Survive

For many years, we have understood that knowledge should not depend on a single individual.

I believe we now need to learn the same lesson about AI.

Context should not depend on a specific chat, a specific account, a specific model, or a specific vendor. A surprisingly large portion of context can actually be structured and preserved if we use the right formats.

There will always be things we cannot fully capture. The tone of a discussion. The uncertainty that existed at the time. The intuition behind certain judgments.

But preserving everything should not be the goal.

The goal should be to understand which parts of the context genuinely require a human being to be present, and to ensure that everything else does not depend on that person being there.