The Unwritten Manual: Why the Most Valuable Knowledge in Your Company Has Never Been Typed

I spent the early years of my career in the mining and construction industries, where the wrong decision at the wrong moment does not cost revenue. It costs lives.

In those years, I learned something that no textbook, no course, no certification program ever taught me. I learned that the most important knowledge in any operation lives inside the heads of the people who have been doing the work for twenty or thirty years. It has never been written down. It has never been documented. It has never been uploaded to an LMS or turned into a PDF. It exists in the way a veteran operator listens to a machine and knows, from vibration, from sound, from feeling, that something is about to fail. It exists in the way an experienced blaster reads the rock face and adjusts charges based on patterns that no manual describes. It exists in the stories told during lunch breaks about the time the ceiling almost came down in tunnel seven, and what the crew did in the forty-five seconds before the bolts were reset.

This knowledge has a name. Researchers call it tacit knowledge. The stuff in your head that has never been articulated, never been codified, never been made explicit.

Dorothy Leonard at Harvard Business School defines it simply: *"stuff in your head that's never been written down."*

And here is the problem that keeps me awake at night: this knowledge is walking out the door. Every single day. And AI, for all its brilliance, cannot replace it because AI was never trained on it.

The $900 Billion Brain Drain

Let me give you the numbers, because I am a business developer and I think in numbers.

Ten thousand baby boomers retire every day in North America. Four million per year. They comprise 31% of the current workforce, and 56% of those retiring hold leadership or senior operational positions. U.S. companies lose an estimated $900 billion annually to turnover-related knowledge loss.

In mining, construction, utilities, and heavy industry, the impact is disproportionately severe. These are not industries where you can Google the answer. The knowledge that matters, how to read a rock formation, when to trust a structural calculation, and when to override it, which shortcuts are safe, and which will kill someone, is accumulated through decades of physical presence in environments that most AI training data has never seen.

By 2030, an estimated 61 million boomers will have exited the workforce entirely. And with them, the unwritten manual, the accumulated operational wisdom of an entire generation, will be gone.

Not archived. Not saved to the cloud. Gone.

What AI Training Gets Wrong

Now let me tell you what frustrates me about the current state of AI in education and training.

The market is exploding. AI in education is a $32 billion industry. Every platform, every startup, every corporate L&D department is producing AI-powered courses. And what are they building?

You give the platform a topic. It generates a course. Modules. Quizzes. Maybe a video script. The content is well-structured, grammatically flawless, and comprehensively generic. It teaches concepts. It does not teach craft.

The data proves the failure: 82% of enterprise leaders say their organization provides AI training, yet 59% still report an AI skills gap. Only 21% see "significant" positive ROI from their AI training investments. Nearly a quarter of leaders say learning paths are not tailored to specific roles. And 50% of employees say high workloads leave "little room" for training, because the training feels disconnected from the work they actually do.

The reason is structural. Current AI training platforms are built on the same architecture: they take explicit, documented, publicly available knowledge and repackage it into digestible formats. They are very good at this. But they can only teach what has been written down.

The veteran mine operator's understanding of how a specific rock type behaves under pressure in a specific geological context at a specific depth, that has never been written down. The construction superintendent's intuition about when a concrete pour is going wrong, based on temperature, humidity, and how the mix looks, has never been uploaded to a learning management system. The safety lessons from an incident that was resolved in ninety seconds by a crew acting on thirty years of collective instinct circulate as oral tradition, not as training modules.

AI cannot teach what it has never been given. And we have never given it the knowledge that matters most.

The Two Knowledge Streams

I think about this in terms of two streams that need to merge but currently flow in opposite directions.

Stream One: AI's Knowledge. Vast, structured, searchable. Every textbook, every standard, every regulation, every research paper, every best practice ever documented. AI knows the building codes. AI knows the OSHA regulations. AI knows the engineering specifications. AI can generate a training module on confined space entry that fully covers every regulatory requirement. What AI cannot generate is the story of what happened to Carlos in shaft 12 when the ventilation system failed and how the crew got him out alive in four minutes using a technique that violates the standard procedure but works because the standard procedure was written by someone who never worked underground.

Carlos is not a real person. But every miner reading this knows someone exactly like him because every mine has a 12-story shaft. The details are invented. The pattern is universal.

Stream Two: Human Tacit Knowledge. Deep, contextual, experiential, and disappearing. It lives in the memories of people who have thirty years of scar tissue, some literal from doing the work. It is the difference between knowing a procedure and knowing why the procedure exists and when it does not apply. It is the knowledge that turns a competent worker into a master, and it is the knowledge that keeps people alive in environments where competence is not enough.

These two streams are not in conflict. They are complementary. AI's strength is breadth, consistency, and availability. Human tacit knowledge's strength is depth, context, and judgment. The question, the business question, the training question, the survival question, is how to merge them.

A Model That Does Not Yet Exist

Here is what I believe needs to be built. Not as a concept. As a product. As a business.

Phase 1: Capture Before They Leave

The first priority is extraction. Before the veteran retires, before the experienced operator takes their buyout, before the master blaster walks out the door on their last shift, we must capture what they know.

This is not a recording session in a conference room. This is structured knowledge elicitation, using AI as the interviewing engine. AI that has been pre-loaded with the technical domain, the standards, the regulations, and the industry's documented knowledge, and then conducts deep, contextual conversations with the expert. Not generic interviews. Targeted conversations that probe the specific knowledge gaps: "The manual says to use a 3-meter bolt pattern in this type of formation. You always use 2.5. Why?"

The AI does not need to understand the answer intuitively. It needs to capture it, structure it, link it to the relevant technical documentation, and make it searchable and retrievable. Modern AI systems can process multimodal inputs, such as voice, video, and observations of actual workflow, to extract expertise from how experienced personnel actually perform their work, not just how they describe it.

Phase 2: Build the Hybrid Course

Once you have both streams, the documented explicit knowledge and the captured tacit knowledge, you can build something that does not exist today: a professional training program that teaches like a master.

Imagine a safety training module for underground mining that does not just list the regulations. It presents the regulation, then presents the story of what happened when the regulation was not enough, told by the person who was there, with the AI providing the technical context for why the standard procedure failed in that specific situation and what the veteran's improvised solution actually accomplished in engineering terms.

This is not a textbook with anecdotes. It is a fundamentally different kind of learning, one where the structured knowledge of AI and the experiential knowledge of humans are woven together into a narrative that teaches judgment, not just compliance.

Phase 3: The Living Knowledge Base

The course is not the end product. The end product is a living knowledge system that continues to learn. New incidents get added. New stories get captured. New operators contribute their own tacit knowledge as they develop it. The AI continuously integrates new data points with existing knowledge, creating a system that becomes more valuable with every shift, every incident, every story.

This is what Shervin Khodabandeh of BCG describes in his research: integrated human-AI systems where AI learns from humans and humans learn from AI, with value scaling as the number of learning interactions increases. The more the system captures, the better it teaches. The better it teaches, the more humans trust it enough to contribute.

Why This Is a Business, Not a Project

I am a business developer. I think about ROI. So let me make the business case.

The cost of knowledge loss is quantifiable. When a senior mining engineer retires, and their replacement makes a geological assessment error that delays a project by three weeks, the cost is often measured in the millions. When a construction superintendent's tacit knowledge about local soil conditions is not transferred, and a foundation requires rework, the cost is measurable. When a safety incident occurs because a new crew lacked the experiential knowledge to recognize a warning sign that every veteran knew instinctively, the cost is not just financial. It is human.

The market is underserved. Current AI training platforms serve K-12 education and university-level learning effectively. Corporate training platforms serve generic professional development. But the intersection of AI-powered training that captures and teaches industry-specific tacit knowledge for operational professionals is almost empty. It is too niche for the big platforms. Too complex for generic AI course generators. And too valuable to ignore.

The pricing model is clear. Mining companies, construction firms, utilities, and heavy-industry operators already spend heavily on training, both for compliance and operational effectiveness. A system that demonstrably reduces knowledge loss and incident rates has a direct, measurable ROI that supports enterprise pricing.

The competitive moat is deep. As the California Management Review argued in March 2026, tacit knowledge is the next competitive moat. The company that captures it first owns something that cannot be replicated because the experts who generated it are retired, and the contextual, experiential knowledge they held is not available from any other source.

What AI Cannot Do, Yet

I want to be clear-eyed about the limitations. AI cannot replicate the physical experience of feeling a machine vibrate in a way that signals failure. It cannot replicate the environmental awareness that comes from spending ten thousand hours underground. It cannot teach instinct.

What AI can do is preserve the articulated form of that instinct, the stories, the explanations, the "here's why I do it this way", and make it available to people who have not yet accumulated those ten thousand hours. It cannot give a junior operator thirty years of experience. But it can give them access to the lessons of those thirty years, structured, contextualized, and searchable, integrated with the formal knowledge they are already learning.

This is the gap. This is the market. And this is what we should be building before the last generation of masters walks out the door and takes the unwritten manual with them.

A Note on Time

There is an urgency to this that the market does not fully appreciate.

The baby boomers are not going to wait for us to build the perfect platform. Ten thousand of them retire every day. Every day that passes without systematic capture of tacit knowledge is a permanent loss, knowledge that took thirty years to accumulate and takes twenty-four hours to disappear.

In the mining industry I came from, we have a concept called the "window of opportunity", a geological term for the brief period when conditions align for extraction. Miss the window, and the resource is not gone, but the cost of reaching it becomes prohibitive.

We are in that window right now. The experts are still here. The AI tools to capture, structure, and teach their knowledge exist. The gap between the two is not technology. It is the decision to act.

The unwritten manual is the most valuable document your company will never have unless you build the system to write it, with the people who know it, before they are gone.

------

Ali Siamaki @Tony Gee and Partners, a business developer with a background in mining and construction operations in Toronto, Canada. His focus is on applying AI to traditional industries, particularly in preserving the operational knowledge that disappears when experienced workers retire. This is the sixth article in the "Singularity Paradox" series.