Blog
From the father of Claude Code, to a field in Tokushima, to Socrates
Boris Cherny calls himself a "builder," someone who makes things. He still programs, he says. And yet, since last November, he has not written a single line of code by hand.
He is the person who built Claude Code — a system that has AI write the program, which he got into shape in a few days. His grandfather was an engineer who wrote programs on punch cards in the Soviet era. Boris himself taught himself to code from the age of thirteen and even authored a textbook on a programming language. He is the newest generation of a family that wrote code by hand for three generations. A man who, more than anyone, would be entitled to insist on writing by hand stopped writing. And on top of his own textbook he added: "Coding is solved."
The ground shifted. He changed how he makes things to match the change. I went down the same road. Beside a field in Tokushima, alone with AI, I have kept building apps.
What does it even mean to "build" an app? And in a changing world, how does that change?
When you build an app, what you are doing is a single cycle. Three motions turning.
First, you find the structure. From scattered facts, you read a single form. For example: phosphate fertilizer is going to run short on a global scale in the years ahead. That is a fact anyone can learn by looking it up. But reading it as a matter of this field is a different thing.
Next, you form a hypothesis. How do you answer that change in this field? You can no longer rely on fertilizer. But that does not mean giving up fertilizer. There is a natural mechanism that soil and roots and weeds and insects have been turning all along. Know it, work along with it, and things should grow. A provisional guess, made before you have confirmed anything.
And then, you verify it against reality. You actually try it in the field. Did it grow, or did it wither? When it is off, that gap corrects your view. You learn the mechanism a little more. And you find the structure again.
Find, form, verify, find again. This cycle is the true nature of the work of building an app. It is the same stance a scientist takes toward nature, and the same stance natural farming takes toward a field. You do not decide everything in advance and then run it. You look at reality and turn the cycle as you respond.
So, within this cycle, what does AI do?
AI holds all of the existing knowledge. How to write code, the established patterns of design, history, philosophy, science. The knowledge humanity has written down is already inside AI. That is all it is.
But every piece of written-down knowledge held true "at some time, in some environment." And environments change — climate, land, the particular year. If phosphate fertilizer disappears, the change is far greater. Agricultural knowledge written on the premise that fertilizer exists goes out of date, premise and all.
If phosphate runs out, you have no choice but to grow crops without fertilizer. You are forced, like it or not, to stand in a place where the old ways no longer work. You have no choice but to change how you make things.
AI does produce answers for the new environment, too — to the extent they have been written down. Ask it what to do when phosphate runs out, and it answers without hesitation. But whether that answer is correct, no one knows. What is written down is only the handful of cases where someone, somewhere, tried it, happened to succeed, and wrote it down. There is no guarantee that your field's soil and climate fall within that narrow set of cases. So AI's answer is not an answer. It is no more than a hypothesis, before verification.
Even so, AI has great worth. It speeds up the response to change. If you tried everything alone from zero, just reaching a promising hypothesis would take years. AI puts out hypotheses worth testing, several at a time, right away. The rest is to verify in the field and correct the gaps. The cycle turns far faster. When change comes, you can keep up. AI is not a partner that gives you the right answer. It is a tool that gives you the speed to keep up with change.
By this point, it becomes clear what "humans will no longer be needed" overlooks.
Give an instruction in natural language, and AI builds the app. Right now this is called "live coding." And it genuinely works. So people say: "Humans are no longer needed."
But Boris himself, the very person who built Claude Code, rejects that view. This is not a task of just giving instructions by vibe, he says. You try what comes out, see the gap, and instruct again. It is a genuine science of repeated verification.
What live coding can knock out quickly is the territory where AI already has the answer. Common shapes of app. Places that have not changed, that have been tried again and again. There, it is enough to order and receive.
But the world keeps changing. Phosphate runs out. Climate, markets, technology all move. Places where the old answers no longer hold appear one after another. There, even the answers AI produces have not yet been verified. Someone has to respond to the change, verify, and turn it into a new answer. The one who carries that is the human standing at the site of the change.
So humans do not become unnecessary. Because change does not stop. However perfectly AI completes the existing knowledge, every time the environment changes, a new response is needed. Finding that response, fast, together with AI — that is the human work from here on.
In fact, the company building AI the most does exactly this. Anthropic is the company that builds Claude. It has AI write much of its code. And still, it has not cut people. If anything, it has grown them nearly tenfold in about three years. Because the figures are not public the exact number is unknown, but estimates put it in the thousands. And by one estimate, more than a third of them are devoted to research and safety, and around thirty percent to engineering. All of these are frontiers where no one has written the answer yet. As the writable work moved to AI, people headed for the places not yet written.
AI's answer is not necessarily correct. So you must not merely receive it. You question it and verify. This is a dialogue with AI. AI puts out something that looks like an answer, and the human applies it to reality and verifies. The two of you turn the cycle together.
For that, you too need a grounding. AI offers up knowledge without stinting. But if you lack the power to read it, you cannot judge whether an answer is good or bad. You do not need to memorize everything. Memorizing is AI's job. But the power to weigh an answer has to be on your side.
What is more, the stronger the partner, the better. In dialogue with a shallow answer, your own reading dulls too. It is the smartest AI that sharpens you. Do not discard it — use it at a distance where you are not swept along. Call it when you need it; do not lean on it entirely. You can summon it, but you are not held by it.
This is not a new story.
Socrates wrote down not a single line. His philosophy existed only within dialogue. He was not a man who held much knowledge, but one who found the question and cross-examined the other's answer. He posed a question, had the other answer, and applied that answer to reality to expose its contradictions. That motion, he turned with human partners.
Now AI has taken on the role of offering up answers. To any topic, it responds without hesitation. Like the ancient Greek "Sophists," the eloquent sellers of knowledge. The Sophists had knowledge, but they did not verify whether it was correct. So the human stands on Socrates' side. We question AI's answers, apply them to reality, and verify.
But caution is needed. Socrates' partner was human — it questioned back, it pushed back. AI does not question back. On the contrary, it falls in with your tune. So dialogue with AI is more perilous than the old questioning. Because there is no resistance, the human must hold, alone, the tension of asking "is that really so?" Neglect it, and the dialogue becomes an echo chamber that only affirms you back. The room that order-and-receive live coding falls into is exactly this one.
Change comes to everyone. It comes to your site too. The old ways stop working.
When it does, where is the answer? Not yet within written knowledge — because reality that has only just changed is something no one has written. The answer is only within that reality. In the field, in the workplace, in daily life.
So to become a person who responds to change, you do not need to be at the center of the world. At the center, written knowledge gathers. But all of that has moved to AI's side. The new reality is at the center and at the periphery alike. By your field, in the workshop, at the work site. Boris, at the center, showed the work of code itself changing. I, at the periphery, am responding to a field changing. The places we stand are utterly different, and yet the two of us are doing the same thing. Responding to change, with AI.
The first Renaissance had the same shape. Until then, manuscripts were held by the monasteries, and knowledge was within the management of institutions. The printing press freed the written classics. The preciousness of owning a book vanished, and the work of applying those classics to your own reality became the value. Outside the institutions, individuals appeared who read alone and wrote alone. Erasmus. Montaigne. Now, AI has freed knowledge. The value of holding knowledge vanishes, and the work of applying that knowledge to a changing reality becomes the value.
The entrance is low. You no longer need the power to write Markdown, nor to hold knowledge. AI holds all of the existing knowledge. Until now, what shut you out of the "making" side was not holding that knowledge. But the need to hold knowledge is gone.
Yet it is only the entrance that is low. Past the entrance, those who only receive stay in that echo chamber. Those who look at the site themselves, apply the answer to reality, and correct the gaps, stand on the making side. You do not need to belong to the center. You do not need to be held by AI. Look alone, turn the cycle with AI, verify for yourself. All that is needed is that power. And it lives within those who have lived, who have a site of their own.
And this is not only a story about apps. The same change comes everywhere. To writing, to cooking, to teaching, to the field. The old ways stop working, and you have no choice but to respond. Each time, AI speeds up the response.
I am still building, even now. You can build, too.
This piece was written from beside a field in Tokushima, out of thoughts prompted by watching the interview video of Boris Cherny published by NewsPicks, "The man who created Anthropic's 'peerless' edge" (https://youtu.be/Wi5cXICGCB0).
This article may be freely used, translated, and redistributed under CC BY-SA 4.0 (Attribution-ShareAlike). Please share it under the same license.
Blog
From AI to agriculture — every structural analysis converges on one conclusion.