Start by Splitting Work into Two Types
Lay out the three libraries side by side and you will find they belong to two fundamentally different worlds.
One is the world of certainty. Work here is fixed, has a correct answer, and cannot afford errors. Accounting is accounting; inventory is inventory; being off by a dollar is not acceptable. The database lives here.
The other is the world of uncertainty. Work here is ambiguous, abstract, and has no single right answer. Things like extracting the judgment locked inside a senior employee's head, helping think through a complex decision, or drafting a strategy. Knowledge bases and rule bases live here.
What these two worlds need is completely different. The world of certainty demands 100% accuracy, identical results every time, and a traceable audit trail. The world of uncertainty needs something good enough, something that sparks useful thinking, and something that gets sharper with use.
First, Get the Three Libraries Straight
To decide where AI fits, start with the three most common types of "libraries." Think of each as a warehouse; the difference is what's stored inside.
Stores structured records that can be queried. It holds data: orders, inventory, customer lists. This is the foundation of the certain world.
Stores organized, contextualized knowledge that can be put to use: a clearly written standard operating procedure, or the accumulated experience of a team.
Stores decision logic: "if this, then that" conditions, one rule at a time. It captures how to judge and decide.
Two quick notes before moving on. First, a rule base is technically a type of knowledge base; it just specializes in decision rules. Second, you will notice ERP is not listed as a library. That is because ERP is not a library. It will show up shortly in a different role.
ERP and Agent: The Signature System of Each World
Each world has one defining player. Putting them side by side is the clearest way to understand what has changed in recent years.
The signature system of the certain world is ERP. It is a full suite for managing company operations, linking finance, inventory, procurement, and HR into one engine that runs on hard-coded rules without deviation. It sits on top of databases. Its job is to be stable and precise, and software has been doing that job very well for decades.
The signature system of the uncertain world is what everyone is talking about now: the Agent, an AI that can judge situations and act on its own. It reads context, makes its own decisions, and adjusts based on what is in front of it. Every response is shaped by the current situation rather than a fixed script. It sits on top of knowledge bases and rule bases.
There is one critical point here. Simply giving an Agent a knowledge base and a rule base is not enough for it to get things done. That material needs to be integrated into discrete, executable capabilities before an Agent can pick it up and run with it. That integrated package is what we call a Skill.
The Skill Library: Turning Three Libraries into Things AI Can Do
This is the most important concept in the whole article. Package everything needed to complete a specific task into one unit, and you have a Skill, often called a Skill Package.
Every Skill Package contains three things:
Historical records and past examples. This kind of data serves as raw material for AI judgment. It is different from transactional data in an ERP, where every record must be exact.
The background story, the reasoning behind approaches, the context needed to understand a situation. This lets AI read the room rather than just follow steps mechanically.
How to judge and what standards to apply. This translates the decision logic inside a senior employee's head into criteria AI can actually reference.
In short, a Skill Package is what you get when you take a database, a knowledge base, and a rule base and integrate all three specifically around one concrete task.
For example: after a customer complaint review meeting, a resulting Skill Package might look like this: historical return records as data, a communication SOP for de-escalating customers as knowledge, and compensation approval thresholds as rules. Hand that package to AI, and next time a complaint comes in, it can help you think through a response based on what is already there.
Many Skill Packages accumulated over time become a Skill Library.
The AI employee analogy makes this concrete. The database is what the employee can look up. The knowledge base is what they understand. The rule base is how they make judgment calls. The Skill Library is the list of things they can actually do. Knowing a lot is nice, but the employee who gets things done is the one who can execute. The Skill Library is how AI moves from knowing to doing.
In the uncertain world, the Agent is the engine. The Skill Library is what it knows how to do. Together, they make an AI employee that can genuinely take work off your plate.
How a Skill Library Grows
The clearest articulation of this idea of distilling professional expertise, workflow, and experience into reusable Skills comes from Anthropic's Claude Skills, released in 2025. The official framing: "Turn your expertise, processes, and best practices into reusable capabilities." In plain terms: do not let a good approach happen only once.
This requires a shift in mindset. When you are in a meeting or doing a piece of work, the goal is not just to finish the task. Ask one more question: what reusable Skills can we extract from this?
An ordinary meeting ends with the conversation dispersed. The knowledge stays in the room, and the next time something similar comes up, everyone starts over. An upgraded meeting ends with a few Skill Packages in hand. Next time, AI or the team can follow them without needing to reinvent the thinking.
I use a simple standard for myself: a genuinely valuable meeting should produce at least three Skills. Treating that as a personal discipline forces vague discussions to converge into specific, repeatable capabilities. It is also a diagnostic. If a meeting cannot yield even one Skill, it usually means the ideas have not been fully worked through yet.
This Is How Tacit Knowledge Becomes an Asset
Taken together, this entire process is about turning the hard-to-articulate experience inside a company into something visible and reusable.
The most valuable capabilities in any company are usually locked inside a few senior people's heads. How a master craftsperson reads a batch of materials. How a veteran salesperson reads a client. The unspoken nuances that make cross-department coordination work. When that person leaves, that capability walks out with them.
Distilling that experience into Skill Packages and accumulating them into a Skill Library transforms it from one person's capability into an organizational asset that anyone can access and that compounds over time. The significance of that goes well beyond saving one person's hours. It means the team's judgment can be preserved and built upon.
Where Can a Company Start
No need to commit large budgets up front. I would suggest three steps.
List the high-frequency tasks in your company that are deeply experience-dependent and have no standard answer. Those are the best candidates for distilling into Skills.
Accounting, inventory management, order fulfillment: tasks that already run fast and accurately should stay with existing systems. Do not disrupt them just to use AI.
Pick one experience-heavy task, and at the next meeting, try to converge the discussion into one or two Skill Packages. Let a small team try them out, then expand once the process runs smoothly. That is how a Skill Library grows: one meeting at a time.
Common Pitfalls
Three pitfalls that come up most often, with what to do about each.
Pitfall 1: Adopting AI for its own sake and grafting it onto workflows that already run fine
Response: First confirm whether the task has a correct answer. If it does, the existing system handles it. Not everything is suited for AI; certain tasks already have software doing them well.
Pitfall 2: Judging AI by a zero-error standard and dismissing it entirely when it slips
Response: For uncertain tasks, the question is whether AI helps the team think more clearly. In the uncertain world, the standard is usefulness, not perfect accuracy every time.
Pitfall 3: Running many meetings and doing a lot of work, yet leaving behind nothing reusable
Response: At the end of each piece of work, ask: what Skills can we extract from this? Make Skill distillation a standard closing step rather than an afterthought.
A Note for Leaders
The emphasis of AI adoption will gradually shift away from speed. Tools are available to everyone, and they are converging fast. Competing purely on velocity quickly becomes a red ocean where everyone looks the same. What creates lasting differentiation is how much judgment and knowledge your organization has accumulated that others cannot simply copy.
Extracting the expertise locked inside people's heads, packaging it into Skills, and growing it into an organizational Skill Library: that is the investment most worth making in the age of AI.
It is also what I do: helping companies and individuals distill tacit knowledge into AI-ready knowledge assets and Skills, so AI can genuinely connect with and extend your judgment.
Two free online talks every month.
Join the LINE Community ↗