Will Schools Become Unnecessary in the Age of AI?

I still do not know whether the age of AI agents will really arrive. But it is clear that tools such as Claude Code and Codex have begun to change how people work in software development. Searching for knowledge, organizing it, writing text, writing code, and fixing code: parts of these intellectual tasks are already being heavily supported by AI.

What, then, will this change bring to learning spaces? When AI can generate personalized materials and exercises, visualize learning progress, and provide coaching when needed, will schools and cram schools become unnecessary?

I think that way of asking the question is a little too rough. What needs major redesign is the conventional school system that gathers students of the same age in the same classroom and manages them with the same textbooks, the same pace, and the same evaluation criteria. By contrast, learning spaces where people can safely fail, interact with others, and refine their own values will become more important.

In this article, I want to separate “school as a system” from “school as a space” in the age of AI agents.

By school system, I mean a mechanism that gathers students of the same age in the same classroom and manages them through the same textbooks, the same pace, and the same evaluation axis. By school space, I mean a place for safely failing, interacting with others, and refining one’s own values.

Why the Existing School System Has Reached Its Limits

Since the internet became widespread, there has been an opinion that apps and web services are enough for learning, that people can watch interesting lessons on YouTube, that classes over Zoom and other online meeting tools are enough, and therefore schools are no longer needed. With the arrival of AI, this opinion has accelerated further.

I do not agree with it. When the discussion becomes a conflict over whether schools are needed or not, school as a system and school as a space get mixed together.

As for the school system, based on my earlier article なぜ教育機関は能動的学習に転換できないのかの７つの仮説 and Ivan Illich’s Deschooling Society, I think it has already reached its limit. Illich’s criticism was that schools are supposed to be places where people learn, but in reality they become institutions that make people believe that learning means going to school. AI will make that limit even more visible.

First, as technology advances, school is no longer the optimal place for acquiring knowledge. At the same time, school as a social institution has a side that no longer asks about knowledge itself, but functions like a business of buying graduation credentials. This has both merits and drawbacks, but education has also been pulled not only into pure scholarship but into the logic of productivity and business, where the question is whether it is useful to society.

Against that background, I think it has become broadly understood that active learning is better. But I have also heard how difficult things are in schools. Even when schools want to introduce personalized learning, they have neither the budget nor the staff. What can be done when one person has to look after dozens of students? People are introducing new technologies despite difficult conditions, but there are also cases where those technologies are being used to maintain the old education system.

I think the better direction is to act in order to gather allies who can aim for change despite the difficulty, and I deeply respect the people who are taking that kind of action.

How AI Changes Knowledge Acquisition

What is striking about today’s AI is not simply that it can solve problems with specific answers, such as university entrance exam questions. In some specialized tasks, such as implementing complex applications and finding bugs, AI is starting to produce results that match or even exceed those of experts.

AI is beginning to show high performance in various kinds of labor that used to be called intellectual. In that sense, the meaning of simple knowledge acquisition is changing. This is a very important premise.

Several years ago, I asked someone in education about an app that uses algorithms to personalize learning for problems like those that appear on university entrance exams. If the technology exists, I wondered whether it should be used somewhere else instead.

The answer I received was that things that absolutely must be learned should first be compressed by technology and learned efficiently, and then the time created by that should be used for creative learning. I was not fully convinced, but I did think there was something to that view.

With the arrival of AI, I think this way of thinking will go further. For example, based on what a student genuinely wants to learn, it becomes possible to create a textbook on the spot that matches the student’s level. Exercises for checking understanding can also be created immediately. Learning status can be visualized, and AI can provide coaching as needed.

When that happens, the need to use the same textbook, sit in the same classroom, and listen to the same lesson at the same pace will decline. There is still the question of how to check progress when everyone learns separately. But it is realistic to imagine teachers understanding the whole picture through AI coaching and system-based visualization, then intervening when individual adjustment is needed.

Why Learning Spaces Become Important

Then are learning spaces such as schools and cram schools unnecessary? No. Precisely because knowledge becomes commoditized, spaces that are safe, allow people to interact with others, secure diversity, and contain a good kind of discomfort become more important.

I think learning spaces have three major roles.

The first is to be a place where people can practice safely. In classroom learning, there is to some extent something like a correct answer. In practice, however, facts are continuous, and often it is not clear what the correct answer is. Failure can lead to success, and the reverse is also true. In practice, giving meaning to events becomes at least as important as choosing the option that looks correct. To learn that, people need a space that is safe and allows experimentation.

The second is to be a place where people encounter discomfort with others. AI has become very smart, but it sometimes produces hallucinations: incorrect information presented as if it were correct. Even when there is no hallucination, depending on the context, AI may list facts that do not work well.

In addition, relying only on a single AI may create an even stronger echo chamber. AI tends to return arguments that feel comfortable to the user. That is why the experience of having one’s thinking challenged, or feeling discomfort through differences with others, becomes important. Such experiences are hard to obtain except in a space where diversity is secured.

The third is to be a place where people refine their values. AI will cause productivity to explode. When that happens, the kind of thinking that prioritizes time performance may change again. There will likely be more opportunities to face the question of what people live for. We need to refine our own values. I think those values can only be refined through contact with others.

Given these points, the meaning of having other people nearby who support you in a physical space, and of feeling a sense of belonging there, is large. If we look only at knowledge acquisition, the necessity of schools may decline. But as places where people practice, interact with others, and refine their own values, learning spaces will become more important.

Conclusion

There are broadly two things we should discuss about learning spaces in the age of AI agents.

One is to design AI-agent-based learning and create a new school system for it. The other is to build, as a community, the learning spaces, physical spaces, and ways of being human that presuppose those changes, while engaging with the real world.

The ideal, I think, is for the learning logs and project experience gained in agile cram schools to connect to inquiry in schools with a stronger public role. This is not about replacing schools with cram schools. Rather, it is about how to connect the agility of cram schools with the public role of schools.

Then, the issues that become visible at school can be explored more deeply in cram schools and local communities, so that learning spaces become connected in ways that allow movement back and forth between them. If we can design that kind of relationship, learning in the age of AI agents will not be confined to mere personalized optimization, but will expand into richer forms of practice.

If we look only at knowledge acquisition, the necessity of schools may decline relatively. But as places where people give meaning to events in practice, accept discomfort with others, and refine their own values, learning spaces will become more important.

That is why we need to separate school as a system from school as a space. I think the discussion has to begin there.

Supplement: Experiments and Research Behind This Article

This article is based on several experiments and pieces of research. The experiments are still continuing, and I am also preparing new ones. The argument may change, but this is a summary of what I think at this point.

One concrete experiment I refer to is the Zenn article AI時代のプロダクトは「固定された成果物」ではなく「可能性の束」になる !? - Cookflowを作った. In that experiment, I used prototyping to show that when developing products with AI agents, there may be a shift from conventional “fixed deliverables” to “bundles of possibility.”

The research behind it is AI時代のプロダクトはプロセスになる on experimental-commons. experimental-commons is a website I made. I have AI agents handle most of the research and first-draft writing for topics I care about, while humans are involved in planning and editing.

Other research from experimental-commons that I refer to includes 身体性、AIエージェント、分断社会, ウィトゲンシュタイン、言語の限界、LLMの限界, and AIエージェント時代の学習空間. The last one is what led me to write this article.

I have also written related articles before, such as 2020年以降、教育業界はどのように変化するか（するべきか） and なぜ教育機関は能動的学習に転換できないのかの７つの仮説.

When I write Zenn articles, I use AI to plan and define specifications, implement from there, look at the result, and then have an AI agent write the first draft of the article. For this web diary, I sometimes use AI for proofreading, such as typo checks, and for automatic translation. For now, though, I write the first draft by hand.