— Toward the Idea of Knowledge Routing —

Recently, I've been hearing the same argument more and more often in software development teams:

"AI can generate documentation now. We don't need static documentation anymore."

It's an understandable conclusion.

Modern generative AI can organize information instantly and present it in a form that is easy for humans to consume.

As a result, the following assumption is becoming increasingly common:

"If we have dynamic documentation, static documentation has become obsolete."

However, this trend concerns me.

The issue is not whether AI is capable.

The issue is that many teams seem to be making this decision without fully recognizing the assumption they are betting on.

Dynamic Documentation Is Not a Complete Replacement

Today, many organizations effectively have two kinds of documentation:

  • Static Documents: knowledge intentionally designed, organized, and maintained by people.
  • Dynamic Documents: knowledge generated on demand by AI.

These are not competing concepts.

They serve different purposes.

Yet the discussion often turns into:

"Since we have dynamic documentation, static documentation is unnecessary."

I believe that conclusion is risky.

AI Can Only Answer the Questions You Ask

The situation reminds me of the difference between television and the Internet.

Television exposes you to information you weren't actively looking for.

As a result, you naturally encounter topics outside your own area of expertise.

The Internet changed that.

Most of what you see is determined by what you search for.

Generative AI pushes this one step further.

AI can answer your questions.

But it cannot reliably answer the questions you never thought to ask.

This is not a limitation unique to AI.

Humans are exactly the same.

We cannot answer questions that are never asked.

That is precisely why we still need something that reveals the overall structure of a domain before the questions even begin.

Static Documents Are Maps

I think of static documentation as a map.

Dynamic documentation generated by AI is more like navigation built on top of that map.

Navigation can guide you to a destination.

But it cannot guide you toward a destination you never knew existed.

This distinction becomes critical during maintenance, operations, and knowledge transfer.

Without the map, teams eventually reach a state where

"We don't even know what we don't know."

For organizations dealing with complex domains, that is a serious risk.

We Are Betting on a Particular Future

Of course, an organization can decide to eliminate static documentation.

But doing so implicitly accepts a very specific assumption:

"Humans will never need to read static documentation again."

If that future truly arrives, perhaps the decision will prove correct.

But what if it doesn't?

Then the organization will have discarded the very map that allows people to understand the domain as a whole.

My concern is that this trade-off is rarely discussed explicitly.

The question isn't whether AI is powerful.

The real question is whether we have enough evidence to justify betting on that future.

Where Do We Go From Here?

If static documentation still has a role, what should that role be in the AI era?

I don't believe this is simply a documentation management problem.

The real challenge is no longer

"How do we store knowledge?"

but rather

"How should knowledge flow between people and AI?"

In the next article, I'd like to explore this idea through a concept I call Knowledge Routing.