It is more meaningful to focus on what we gain from AI than to dwell on what we might lose.

Read the original article (in Japanese):

Prologue: You Don't Need a Wood-Fire Stove to Use a Rice Cooker

A company banned its new engineers from using AI. The reason: AI-generated code had accumulated over 800 issues that the engineer couldn't understand at all. But was AI really the problem? No. What was needed was simply teaching the correct way to use it. Banning AI because it makes mistakes is nothing more than the easy way out for those responsible for training.

Do you need experience with a wood-fire stove to use a rice cooker? Most people today have never cooked rice over an open fire — yet nobody fails with a rice cooker. If anything, more people cook rice more easily than ever before. What's needed is knowing how to use the tool correctly.

What management needs going forward is not to ban AI, but to build work processes and education around the assumption that AI is part of society.

Chapter 1: AI Is a High-Performing Subordinate

A capable subordinate works fast. Give them instructions and they deliver. But that doesn't mean you can hand everything over blindly. The one who signs off on the output is the manager — and only someone who understands the work can take responsibility for it.

AI works the same way. It lies without hesitation, weaving inaccurate information into plausible-sounding text, producing code that looks functional but is fundamentally broken. Treating AI output as gospel — that's exactly what led to 800 issues going unnoticed.

AI is a tool for capable people to work smarter, not a helper robot that delivers results when incompetent people dump everything on it.

This cuts to the heart of AI's asymmetry: it accelerates those with foundational knowledge, and exposes the emptiness of those without. That asymmetric force is the true nature of AI as a tool.

To manage a subordinate effectively, you must understand the work well enough to give proper direction and correction. The same applies to AI. Foundational knowledge and the ability to verify output are prerequisites. The right approach is not to ban AI, but to teach people how to use it.

Chapter 2: When the Tool Changes, Society Changes

Humanity has always evolved alongside its tools. From caves to houses, then toilets, baths, and kitchens — cities and civilizations were built on each successive foundation. The automobile diminished walking but created logistics networks, new economic zones, and the concept of suburbia. The internet may have weakened memory and handwriting, but it democratized information and transformed business models.

Every time, there were those who said: "We'll lose the ability to walk." "We'll lose the capacity to think." "We'll lose the ability to calculate." And every time, the underlying base of society shifted anyway. Those who refused to adapt lost their place in what came next.

AI is the same kind of tool. The ability to generate language independently may decline somewhat — but in return, humans will be able to think more deeply and carry their ideas further. Denying the shift means forfeiting everything built on top of it. This is not a question about AI. It is the structure of civilizational progress itself.

Chapter 3: Will Capable People Work Smarter, Leap Further, or Both?

"Anyone can produce great work with AI." That's a serious misunderstanding. Generative AI is designed to produce answers users will be pleased with. Even if A is correct, it may offer B if it senses the user would prefer it. If you can catch that and correct it, fine — but without that ability, errors simply pile up.

Using a smartphone's autocorrect doesn't mean you no longer need to learn kanji. The difference between words that sound alike depends on context and judgment. Knowledge is what lets you catch mistakes. The same is true with AI — without foundational knowledge, you'll simply be played by it.

There are three prerequisites for effective AI use:

Foundational knowledge in your field

The ability to verify AI output

The judgment to say "that's wrong"

With these in place, AI becomes a powerful weapon. A four-hour task that once required one hour of thinking and three hours of execution can become three hours of thinking and one hour of verification. The result: far more time devoted to actual thought. That is what it means to say AI makes capable people more capable.

Chapter 4: Judgment and Responsibility

AI replaces "processing" — so we've been told. But as that processing deepens, "judgment" is increasingly entering AI's domain too. Information gathering, pattern recognition, cross-referencing past cases, drawing conclusions — these are all tasks at which AI excels. Sentencing in minor court cases, medical diagnosis — at their core, these are exercises in deriving optimal answers from vast datasets. The day when AI's learning surpasses human intuition is not far off.

So what remains for humans? The act of owning the outcome. When AI outputs "this diagnosis has the highest probability," it is a human who must deliver that news, walk alongside the patient's life, and face whatever comes. In court, it is a human who must receive the full weight of the defendant's circumstances.

Generating judgment moves to AI. Owning that judgment stays with humans. And to own it, you must understand the basis on which AI made its call. To bear responsibility is to know what you are responsible for.

Chapter 5: AI Nativity Will Widen the Gap Exponentially

Banning AI for new employees doesn't protect them. AI-driven society is already in motion — it cannot be stopped. All a ban does is leave people behind.

Employees at AI-adopting companies grow more productive every day. Those at AI-banning companies stay at the same pace. In a few years, that gap will be insurmountable. "AI-native" will become the standard for the next generation. Where gaps were once created by differences in knowledge and processing speed, they will now be driven exponentially by whether or not someone can use AI effectively.

What matters is accumulating experience — including making mistakes — in a safe environment, and doing so early. Banning bicycles because children might fall only produces adults who can't ride. Teaching how to fall, how to handle risk, and how to recover — that is education. Teaching people to use AI, including how to spot its errors and risks, is the core of education going forward.

Chapter 6: Redefining What It Means to Be "Good at Your Job"

In the age of AI, the definition of competence is changing. Speed and knowledge once defined excellence. What will define it going forward?

Domain	Before	After
Thinking	Think from scratch and produce answers	Design the right questions and critically evaluate and develop AI outputs
Judgment	Rely on experience, intuition, and memory	Make final decisions and take responsibility based on AI-generated data
Validation	Notice issues through one’s own knowledge and experience	Require the knowledge to verify AI outputs as a prerequisite
Execution	Perform the entire process independently	Delegate processing to AI and focus on directing, reviewing, and refining
Expertise	Strength lies in knowledge volume and processing speed	Strength lies in the ability to judge correctness and understand context of AI outputs
Inequality	Gaps in knowledge and processing ability	The gap expands exponentially based on the ability to leverage AI effectively

Across every profession, the ability to effectively deploy AI as a capable subordinate will determine the quality of one's work. For management, this also means rethinking how performance is evaluated.

Epilogue: Bringing Your Own Quality to AI

When a creative writer uses AI, what happens? AI produces the material, builds the skeleton. But it only becomes a work of art when the writer's own experience, emotion, perspective, and choice of words are layered onto it. Submitting AI output as-is is like assembling a model kit and calling it your own creation.

Does the programmer bring their own architectural thinking to the code? Does the marketer bring their own understanding of the customer? Does the researcher bring the question only they could ask? Whether "you" are present in the AI's output — that is the dividing line of value going forward.

The definition of incompetence is also changing. It used to mean slow processing and limited expertise. Going forward, it will mean being unable to add yourself to AI's output, and being unable to catch AI's mistakes. "The programmer who let AI write the code and ended up with a mess." "The student who submitted an AI-written report as their own." These are the defining examples of people who will not be able to work in an AI world.

The tools have become state-of-the-art. What is being tested now is the person holding them.

Read in Japanese ↓（For Japanese learners!）↓

このブログを検索

CRIA Global – Japan Workplace Knowledge for All Workers