Since I started learning programming, I’ve been using Emacs, so it was natural for me to write and code in the same environment. From time to time, editors suitable for writing would appear, and I tried them, but they never felt quite right.

Around 2018 (?), I switched from Emacs to VS Code, and I started writing blog posts and other content in VS Code as well. Then, around 2023, GitHub Copilot appeared. Press Enter, and it intelligently suggests code that looks just right. It autocompletes variable names and function names with ease—it was a revolutionary convenience.

I think it was initially only effective for programming languages, but soon it started autocompleting documentation as well. It would generate suggestions in proper Japanese, seemingly understanding my intent. Wonderful!

But that lasted only until recently. Regrettably, I’ve disabled GitHub Copilot for Markdown files.

The reason is that I’ve increasingly felt that generative AI was pulling my thought process in its direction. I think everyone has their own way of writing, but in my case, I write more intuitively. As I gradually write, my thoughts become clearer and more articulate, eventually leading to a conclusion.

With this style of writing, when a long suggestion appears and I accept it, my own thinking gets dragged in that direction. After accepting suggestions a few times, what I end up with is “something” completely different from what I originally wanted to write.

Once I realized this, I understood that letting GitHub Copilot write for me wasn’t working. If I don’t create that initial spark myself, the result feels like a bland, lifeless piece of text.

Of course, writing manually takes time, and there might be incoherent parts or typos. In such cases, it might be fine to let AI handle minimal editing.

Text Suitable for AI

If there are documents that are suitable for AI creation, I think READMEs and documentation for software that’s already been developed to some extent are good candidates. Since the underlying software code exists, having AI comprehensively examine it and create documentation seems like a good fit.

Blogs, on the other hand, are different—they’re a place for me to express concepts that exist within me. Having AI involved in that feels somewhat unsettling.

Translation Between English and Japanese

Recently, there was a controversy on zenn.dev about an automatic English translation feature for blog posts.

Having it enabled by default might not have been the best choice, but I think automatic translation itself is fine. There are many articles in Japanese with excellent content that don’t reach a global audience simply because they’re in Japanese.

For example, when I want to share a Japanese article about CodeRabbit, it’s a hassle to translate it into English before sharing it on Slack. I want to showcase the great engineers in Japan, but it’s a real shame that the Japanese language becomes a barrier.

Of course, there might be mistranslations, and someone might suddenly point them out, but isn’t that all part of communication? If there are mistranslations, it’s not my fault—it’s an issue on zenn.dev’s side. It’s a waste to avoid global outreach because of fear of 0.001% mistranslations.

Conclusion

So, this text was delivered with 0% AI flavor (only translation by Claude Code :-)).