Why We Need to Rethink Knowledge Management (KM) in the Age of AI ?

By Sébastien Wiertz, May 7, 2025

November 2003. I was heading to Guadeloupe for a break and realized, a bit late, that I had forgotten my books entirely. Using my laptop and a painfully slow modem connection, I ordered “Enterprise Information Portals and Knowledge Management” by J. Firestone from Amazon. No Prime back then, and shipping alone cost me eleven euros. Not exactly a holiday reading, but it got me thinking. It was my first real taste of knowledge management, and it ignited quite a passion for the subject. Since then, I’ve read numerous books, implemented projects, given presentations and speeches, shared thoughts in various communities (tribes), and watched the field evolve. From knowledge management to enterprise 2.0, to the early conversations around web 3.0 and the semantic web, the digital workplace, and now AI. I’ve seen theories applied, misapplied, and sometimes rediscovered. Today, I find myself returning to KM with renewed attention.

Not out of nostalgia, but because now I believe we finally have the tools that were missing in the past. With my accumulated experience in KM and AI, I am diving back into this subject ready to turn ideas into practice and make knowledge work at last.

I believe what we hoped for in the past can finally work out now. So let’s dig into the subject again , this time for real.

The Rise and Fall of Knowledge Management

In the late 1990s and early 2000s, Knowledge Management (KM) was everywhere. Organizations had KM programs, Chief Knowledge Officers, and grand plans to map expertise across the company. The SECI model by Nonaka and Takeuchi gave language to the growing realization — that knowledge is dynamic, and tacit knowledge plays a critical role. As I explored in my 2008 article “Why Should I Share?“, the SECI model defined four stages of knowledge conversion: model outlined four stages of knowledge conversion:

• Socialization (tacit to tacit)
• Externalization (tacit to explicit),
• Combination (explicit to explicit),
• and Internalization (explicit to tacit).

It emphasized the dynamic interaction between tacit and explicit knowledge within organizations. Though it may seem abstract at first, it accurately reflects how knowledge flows among people, teams, and systems. In practice, this model proved highly actionable something I experienced firsthand.

This interplay between theory and practice soon became tangible. As organizations expanded globally, the center of gravity began shifting away from the headquarters. A growing number of employees were now outside our borders, which made it increasingly difficult to concentrate knowledge in a single location or rely on proximity and face-to-face exchanges. At that time, the rise of intranets appeared as the perfect solution. These new platforms became essential to communicate efficiently across regions and to store crucial information. the objective was clear: make embedded knowledge, such as supplier processes and lessons learned, visible and accessible beyond local offices and traditional networks. I vividly recall explaining taxonomy to a senior VP in procurement to support this goal. It wasn’t just about invoices or pricing; it was about mapping expertise and defining a shared technical language.

At that time, supported by visionary leaders, we built a domain expertise map and standardized knowledge to make people and information findable. While cultural barriers made sharing explicit knowledge difficult in forums, this approach significantly improved knowledge visibility and access across the organization.It wasn’t just about theory. We acted. The CTO joined and together we built a map of who knew what(Technically). Buyers and technologists were indexed by domain expertise, allowing quick identification of internal experts and product-related information. This effort required designing a proper taxonomy, a shared technical dictionary, supplier classifications, and lists of relevant disciplines (among others). Despite this success, some challenges persisted. In internal forums, for example, cultural barriers made people hesitant to share openly. Fear of blame often prevented experts from expressing their knowledge explicitly. Still, overall, the system made expertise visible and accessible. The process of sharing was initiated in the organization and you knew who to call to exchange on a specific expertise.

Then, slowly, it all faded somehow. There was a shift in mindset where digital platforms began to take precedence over human interactions. A first paradigm shift appeared in 2007 with LinkedIn. People began to maintain and share their competencies outside the walls of the company. The question of “who knows what” had moved to external networks. Companies no longer had their hands on who knows what. Systems designed to externalise expertise ( Tacit -> Explicit) often felt like extra work. Engineers, in particular, were reluctant to write down their know-how in formal documents. The culture was one of “knowledge is power,” and protecting what you knew gave you leverage. Balancing openness with this defensive mindset was always a challenge (and it still is today).  By 2012, I was already writing about the decay of KM structures and the loss of shared memory. In an article, I explored how the disappearance of taxonomies, directories, and mapped expertise was leaving organizations blind to their own knowledge. It wasn’t just a technical failure; it was a cultural one. The mechanisms to surface who knows what had eroded, and with them, the trust that knowledge would be found and valued. We saw the signals. We moved on anyway. Many organizations were dreaming about the next wave of technology that could magically resolve this persistent issue. Knowledge Management was quietly stored away, waiting for better days. Seen as old and outdated, it had been left behind in favor of trendier market hypes.

Technology Is No Longer the Differentiator

Over the past two decades, companies have jumped from one wave of technology to the next. Social platforms, cloud systems, collaboration tools. But what really separates one engineering company from another today isn’t the tech stack. It’s the way they manage and activate their knowledge. AI, cloud, and data platforms are accessible to everyone now. The true advantage lies in the ability to capture expert insight, share it across the organization, and use it to make better decisions, faster. In short, the winners will be the companies that know how to use what they know.

Now we’re in a new era. The era of AI (Augmented Intelligence). And once again, we’re being told this is the game-changer. And it is, if we use it right. AI, especially large language models and generative tools, has huge potential to amplify knowledge management. But there’s a catch. AI can’t do much with fragmented or tacit knowledge. It needs structured, well-articulated input to be effective. And that brings us right back to the core problem of knowledge management. how to convert tacit knowledge into something usable and shareable ? (Explicit K externalisation)

Tacit knowledge has always been at the heart of the problem. So much of what makes organizations work, especially in engineering, isn’t written down. It’s intuition, rules of thumb, experience accumulated over years. That’s exactly why the SECI model gave such weight to externalization (the conversion of tacit into explicit knowledge). It is the most difficult step in the model, and the one where many initiatives failed in the past.
But  generative AI offers something new. Not magic, not automation, but support. These tools can help experts think out loud, refine their ideas, and articulate what they know in ways that feel natural. They can help transform that deep, internal expertise into something structured and usable. AI becomes a mirror, a sparring partner, a guide. Not a replacement, but a multiplier. A force that helps scale individual expertise into organizational capability.

Beware ! It (AI) can’t create knowledge ! (feel free to argue with me on that ;) ), but it can help extract and structure it. Often in ways that are easier and more natural than traditional documentation.
But let’s be clear (again). AI isn’t here to replace human expertise. The fear that AI will somehow steal or replace what we know is unfounded. In fact, the opposite is true. For AI to be useful, it needs us. It needs our intuition, our context, our judgment. That’s especially true in engineering, where decisions are rarely black and white, and where creative problem-solving often depends on a gut feeling or years of hands-on experience.

Tacit knowledge still plays a huge role. The difference today is that we finally have tools that can help us work with it, without oversimplifying or losing its richness.

Great we have a solution now …. But we also have a problem.

Over the past decade, we have undergone multiple migrations, adopted new platforms, tested social applications like Jive and Yammer, and updated intranets with the hope of keeping pace. Yet each change brought unintended consequences. Knowledge was scattered, hidden, or simply left behind during these transitions. Often, those who knew where important information was stored were not involved in the migration process or had already retired. Their insights on the location and structure of critical knowledge were lost. Silos reappeared, metadata was stripped away, and context vanished. We are now sitting on a goldmine of fragmented knowledge, but without reconnecting and revitalizing this lost information, AI alone cannot help us. To truly make a difference, we need to recover and re-integrate this forgotten knowledge into our organizational fabric.

That’s why the conversation about knowledge management needs to come back, urgently. Not as a dusty old discipline from the early 2000s, but as a strategic capability that powers the AI era. Without strong KM practices, without a real focus on capturing, curating, and connecting knowledge, AI will always be running without substance, without access to the organization’s true intelligence. It will lack the essential input needed to generate relevant and precise support for decision-making and operations.

Revisiting Firestone with Today’s Tools

Revisiting this book now, two things strike me. First, it is always worth re-reading a book after twenty years. Experience adds perspective. You understand things you missed before. Second, back in 2003, Joseph Firestone was already writing about intelligent agents, knowledge life cycles, and the architectural requirements for KM to scale across the enterprise. What was theoretical back then is now at the core of how we think about AI in 2025. Firestone was not just early. He was remarkably forward-thinking. He already described concepts that resemble today’s generative AI tools. He warned that without integration, portals would become just another silo. He insisted that codification was not about control, but about clarity. At the time, I was not ready to grasp the full depth of his thinking. Today, it resonates in a different way.

Firestone imagined systems that could support knowledge production and validation, not just storage. He introduced two key constructs that are worth revisiting today. First, Artificial Knowledge Management Systems, or AKMS. These are automated systems designed to facilitate not only knowledge access but its creation, validation, and integration into the organization. Second, Enterprise Knowledge Portals, or EKPs. These go beyond classic intranet portals, embedding KM directly into workflows, decision-making, and collaboration spaces. At the time, these ideas seemed ambitious. Today, they feel necessary. He separated knowledge into subjective, codified, and contextual layers. He talked about feedback loops, not repositories.

With the emergence of large language models and generative AI, we now have the opportunity to bring these ideas to life. When paired with Retrieval-Augmented Generation, or RAG, we can start building AKMS and EKPs in practice. RAG enables us to ground generative outputs in trusted, contextual data. It brings the best of both worlds: the conversational power of AI and the precision of curated knowledge. This means we can move beyond static intranets and brittle taxonomies toward dynamic systems that evolve with how people work and learn. The technology caught up. Now it’s our turn to catch up with it, and to build what Firestone outlined with focus and intent.

From Theory to Practice

The risk now is to pretend we need to start over. To think AI will magically fix KM. It won’t. AI is not a KM system. It’s a tool that can support KM, if we give it structure, context, and purpose.
That’s why we need to act. Not in grand programs, but in focused moves. One use case. One broken process. One area where expertise is buried or leaking. Start there. Use generative AI not to replace human insight, but to help capture and clarify it.

Take this example. Your company delivers complex equipment. The manuals are buried in a shared drive. No one reads them unless something breaks. Now imagine training a GPT model on those documents. An engineer types, “How do I reset the system after a pressure drop?” and gets a precise, validated response. And if the answer isn’t there? That’s even better. It shows the gap. You bring in the expert. Fill the hole. Close the loop. That’s what Firestone meant by intelligent systems.

We don’t need to reinvent KM. We need to finish what we started.

So here we are. Twenty years after that slow modem in Guadeloupe. And the same questions still apply.
How do we make knowledge visible? How do we ensure it flows, improves, and gets reused?
The answers haven’t changed much. But the tools have. Today, we can do what was once only imagined. But only if we treat KM not as a relic, but as the foundation for intelligent work. Theories like Firestone’s don’t need rewriting. They need implementing. ( this article is the introduction to a series of article that will highlight some potential use cases in AI and KM but also guide you (back ) in the forgotten world of KM.
Let’s not watch this moment pass. Let’s connect the dots we already had. And let’s do it with the tools we finally have at hand.

Because in the age of AI, knowledge is fuel. And it’s up to us to make sure there’s something in the tank.