The struggles of three years ago became today's foundation

Around 2022, I took on my first website development challenge. ChatGPT 3.5 had just been released, and as a programming beginner, I engaged in dialogue with AI while groping my way forward—it was literally a continuous struggle. The technology stack I adopted at the time was as follows: Frontend: Bootstrap 5. Backend: Django (Python). Database: PostgreSQL. Storage: AWS S3.

I still remember the joy I felt when I somehow managed to create a site with blog functionality by combining these technologies. Then, three years passed, and the Claude Code revolution erupted. Well, what should I try with this? When I thought about it, what came to mind was the website multilingualization that I had once attempted but abandoned. Maybe I could do it now. With Claude Code, perhaps I could complete it in about a week. However, now that the implementation is actually complete, I'm experiencing a strange sensation that mixes achievement and fear.

A dynamic system that goes beyond mere "static file translation"

What I aimed for with this multilingualization was not simply "translating fixed pages." Past articles written in Japanese would be automatically converted to English by AI and reflected on the English site. Furthermore, when I write new articles in Japanese in the future, English versions would be automatically generated and published—I aimed to build such a dynamically complete system. Specifically, I combined Django "signals" with a custom translation pipeline to generate English content through APIs when saving articles, implementing a model design that manages data for each language. What's remarkable is that all these procedures—complex backend processing, database schema design, security structure assurance, and deployment to production environment—could be completed through natural language interactions with Claude Code alone. It took only half a day.

Compared to my experience three years ago when I was struggling with ChatGPT 3.5 while building the foundation (as a complete beginner, I invested an entire Golden Week plus an additional two weeks), the development speed has accelerated to a point where "100 times faster" is not an exaggeration. Back then, what consumed most of my time as a beginner wasn't the coding itself, but rather "design decisions." "How should I structure the processing flow?" "What should the database table structure be?" "How should I protect security in the production environment?" Each of these decisions required enormous research and trial and error. And above all, the complexity of migration procedures to the production environment and the endless errors and debugging.

However, with this "Vibe Coding" session with Claude Code, I was able to progress through these processes seamlessly within our dialogue. When I communicated "I want to implement this kind of functionality," the AI provided everything from design proposals to implementation, debugging, and even deployment procedures as a companion. It functions not as a mere code generator, but as a partner that shares risks and design philosophy—this difference creates overwhelming speed. And most importantly, the AI conducts testing and debugging almost automatically through trial and error until success.

Conviction that AI will change society, and frank terror

Through this half-day experience, my premonition that "AI will fundamentally change society" has transformed into stronger conviction. The speed at which technical barriers that once required specialized knowledge and years of experience are being dismantled is beyond imagination. At the same time, I genuinely felt a certain "terror." Such advanced system construction can be realized in just a few hours by someone not accustomed to development. Honestly, fear comes first when faced with such rapid progress.

What remains in the end is human "passion," "determination," and the resolve to "take risks and accept responsibility."

However, upon calm reflection, I believe this implementation wouldn't have been possible without the foundation of "struggling" from three years ago. Understanding how Django works and how it integrates with databases and S3—because I struggled to understand that structure three years ago, I was able to dialogue efficiently with AI and judge the merit of proposals. And above all, it seems that everything stemmed from that strange passion I had three years ago: "I absolutely want to try creating a website myself."

Mathematician Professor Masahiko Fujiwara stated in his book "The Dignity of a Nation" that "the starting point of logic is not determined by logic." AI is the ultimate logical genius with the power to derive optimal solutions. However, determining "what to ask" and "where we want to go"—these starting points—are not determined by logic.

The will to "deliver my words to people around the world." The determination to take risks, accept responsibility, and decide "I will go here." These are jobs that, for now, only humans can do. AI becomes powerful wings that transform those decisions into reality at unprecedented speed. However, deciding in which direction to flap those wings remains our own responsibility, and I feel this more strongly than ever.