Why AI Has Started Learning from Artisans

How to inherit human tacit knowledge

When looking at popular GitHub repositories,
I sometimes get a strange feeling.

AIエージェント。MCP。ロボット。
音楽制作ソフトの拡張機能。ナレッジベース。

At first glance, everyone seems to be doing different things.

However, when tracing the origins of the technology,
it seems they are all heading in the same direction.

That is the theme of "How to inherit human tacit knowledge"

that theme.

Knowledge and tacit knowledge are different

Knowledge can be written in books. It can be turned into manuals. It can be searched.

For example,

WordPressの設定方法

Reactの基本構文

Ableton Liveの操作方法

These things are knowledge. But in this world,
there are abilities that are difficult to explain.

熟練営業が商談の空気を読む感覚。

ベテラン編集者が「何か違う」と感じる違和感。

職人が材料を見ただけで状態を判断する感覚。

漫画家がコマ割りのリズムを決める感覚。

These are hard to turn into manuals. That is why they are called "tacit knowledge."

AI first learned to "see"

Early AI began by recognizing the world.

It was image recognition. Technologies like YOLO

車

人

信号

can find things.
In other words, it learned "what is in the world."

Next, it learned to "move"

In robotics research, studies on imitating human movement progressed.

For example, folding a towel.

It is a simple task for humans, but a difficult one for robots.

Researchers,

手の動き

力の加減

順番

was trained. This is a field called "imitation learning."
AI is not just watching, but also "how to move"

has started to learn.

And now, it is starting to learn "judgment."

Current AI has taken it a step further.

What it is trying to learn is "why it was done that way."

For example, programmers.

Even for code that creates the same functionality, the design differs between a veteran and a beginner.

Music producers are the same.

Even if they use the same software, the finished work is different. The difference is not knowledge. It is judgment.

In the AI industry, there is now a rapidly growing movement to incorporate that judgment.

Changes happening on GitHub

Looking at recent popular repositories, things look a bit different than they used to.

Previously,

完成したソフトウェア

ライブラリ

プラグイン

was the focus.

But recently,

AI Agent
MCP
Workflow
Knowledge Base
Agent Skill

projects like this are standing out.

What they have in common is that they are trying to reuse "how to think" rather than the deliverables themselves.

In the Agent Skill concept, expert knowledge and procedures are saved in a format like "SKILL.md," and AI agents load them only when necessary. It is an idea closer to sharing expertise than sharing software.

What I saw from Ableton Live extensions

A GitHub project I saw recently was interesting.

Ableton Live Extensions Agent Skill
ableton-extensions-skill GitHub repository

This project compiles development know-how using Ableton Live's new
Extensions SDK into a Skill for AI agents.
The author themselves

What is important here is that the AI is not being retrained.

Instead,

プロジェクトの作り方
SDKの使い方
APIの扱い方
ビルド方法
配布方法

knowledge is organized so that the
AI can refer to it only when needed. What is stored is not the code itself.

What is stored is

—in other words, the "decision-making process."

This is exactly the change I have been feeling while looking at GitHub recently. From an era of sharing software to an era of sharing the criteria for making software.

Such a trend is beginning to emerge little by little.

Why now?

In fact, Ableton itself released a new Extensions SDK in 2026. Immediately after an SDK is released, AI does not yet have sufficient knowledge.

Therefore, the idea of "compiling knowledge in this field as a Skill and passing it to the AI"

naturally arises. This is not just about Ableton.

Every time a new technology appears, a trend emerges to convert expert knowledge into a form that AI can utilize.

Humanity has been digitizing knowledge

Looking back, Wikipedia was the digitization of knowledge.

GitHub was the digitization of code. Stack Overflow was the digitization of problem-solving. And now, AI is moving beyond that. What it is trying to digitize is not knowledge, but judgment.

The next resource might be tacit knowledge

Within companies, there is a vast amount of wisdom that exists only in the minds of veterans who have worked there for many years.

Much of it has not yet been sufficiently articulated. If AI can learn and inherit this, the tacit knowledge dormant around the world may become a new resource.

A journey to find the source

Everyone seems to be doing as they please. But if you trace it back to the source, it all connects to the same river.

In the end, aren't they all heading toward the question of "how to inherit human expert judgment"?

And small experiments like Ableton Extensions Agent Skill seem to be getting a glimpse of that future a little early.

#LivingWithAI #InheritanceOfTacitKnowledge #MangaProduction
#Design #Creative #TheWayOfTools