A blog renewal after three years
I renewed my blog with Claude Code and Codex after leaving it mostly untouched since late 2023.
AI really is remarkable. It refreshed the old blog quickly and with a level of polish I didn't expect. The result is honestly faster, cleaner, and more complete than what I could have built on my own.
My workflow was simple: plan with Claude Code, implement with Codex, then have Claude Code review the result again.
Because the original codebase already existed, I could keep iterating on the stack and other details on top of it. Deployment also went smoothly because the AI handles Vercel and Cloudflare well.
Self-reflection
I often feel that I'm missing a lot of theoretical knowledge of development. The way I've worked so far has been about solving frequently changing requirements under tight schedules, with a small team responsible for building and operating multiple products.
To work that way, I couldn't study theory in advance. I had to understand the requirements and problems first, then quickly pick a fitting technology and solution to get the work done.
I could have studied after work, on weekends, or during breaks, but honestly it wasn't easy. Because I never properly rested, my health started to suffer. Balancing work and university meant sitting for long hours, so my lower back started hurting. With so many things going on, I got tired often, and uveitis flared up several times, so I ended up visiting the eye clinic frequently for a few years. These are excuses, of course, but as my personal study time shrank, I started to feel the gaps more clearly.
After developing and studying with AI, though, I'm able to build in less time and pick up more material than I could on my own.
A good example is Claude Code's plan mode. It recommends multiple solutions for a requirement depending on context. Seeing those options has been a great way to learn concepts and understand what makes a good choice.
I think one of AI's biggest strengths is reducing the cost of figuring out what to study, so I can keep building what I want while quickly picking up new information along the way. It's also clearly faster and more accurate than my own coding. I was surprised that it runs tests and points out things I hadn't considered.
The way I used to learn—running into problems while building and then digging through various sources—has shifted thanks to fast AI-assisted development.
One of the developers I personally respect, Kang Dongyoon, wrote The Developer in the AI Era. Along with the point that "the ability to learn and code with AI is very important," there's also this:
If you don't understand AI's explanation, paste it into another chatbot and ask which concepts to study to understand it.
So the flow becomes: I explain what I want to build to the AI, then let it run with the implementation. If Claude's plan mode or Codex's planning mode produces a plan, I look through it, find concepts I haven't seen or don't fully understand, and ask another AI to walk me through them.
Putting the old and new flows side by side:
Old study flow
- Just start building.
- Run into trouble.
- Learn naturally through trial and error, or discover new concepts I want to study.
- Search official docs and blogs.
- Apply what I learned to the existing work and refactor to improve it.
Study flow with AI
- Plan or implement with AI.
- Use the plan or output to surface things I didn't know or want to study.
- Ask AI to explain those concepts with examples.
Even just from comparing the two lists, the number of study steps has clearly dropped.
There's still trial and error with AI, of course. But the amount of trial and error feels much smaller than when I work alone. And since AI catches most issues through tests, I haven't run into anything critical so far.
The most obvious benefit of AI is development speed and the convenience that comes with it. But whether I write the code myself or use AI to help, the biggest benefit is that AI can test the result quickly and from many angles.
After AI finishes development and testing, reading a summary of the result and confirming it matches what I wanted is much more efficient in terms of both time and cost.
My thoughts
AI makes some raw knowledge less scarce, but it makes judgment more important. It sounds paradoxical at first, but it's a bit like the difference between knowledge and wisdom.
Knowledge feels increasingly replaceable by AI, and what matters more is the judgment to use AI well. That doesn't mean knowledge isn't needed at all—without enough foundation, it's hard to tell whether an AI-generated answer is actually any good.
I don't know how heavily other people are using AI yet, but I plan to use it more and study with it more seriously. The world is changing fast, and I want to try many things with AI—not only development, but other areas too—learning and challenging myself so I can keep up with that pace.