My Thoughts on AI-Assisted Content Creation

I have visually verified the links introduced. I use GPT for drafting and research. Also, since this article is just me expressing my thoughts, it is rambling and you won't get any good information out of it. You don't need to read it.

https://info.zenn.dev/2026-03-10-ai-contents-guideline

Guidelines were recently announced here on Zenn, so I thought I would take the opportunity to express my feelings. Are these really just feelings?

As for me, I have been using AI for content writing from the beginning, as you can see. However, I am always careful about whether it is sincere to the reader to just churn out articles instantly, or whether I am providing wisdom or knowledge that offers a return on the time spent reading this article using generative AI. That is why I sometimes note, This part has not been edited by generative AI.

There are three reasons for this.
First: If I were to end with content like "Generative AI said this!" based solely on research using generative AI, I might as well just post a shared link to ChatGPT or similar and be done with it. There would be no room for me to get involved. In fact, it would be a case of "Don't put it through me via filtering." Agents with deep research capabilities, web search, or skills can acquire a certain amount of the latest information just like news feeds, and I assume that those who are streaming AI news every day are writing their articles with considerable labor savings. I don't actually know if that's true.

Second: It is because I want you to read only the good parts through my editing and my insights. The world of generative AI is said to progress in one month what takes other industries a year to achieve. I have been touching it since the days of GPT-3.5 (December 2022), and since the time GPT-4 was released, I have been thinking quite far ahead, not only with code generation but also with implementing automated test design generation. However, I still have a desire to have those who have not yet had a chance to touch generative AI, those who are not brave enough, or those who underestimate generative AI, actually touch and try it out. Therefore, I believe it is fine if you can take away some knowledge by tracing my thoughts based on the expertise I have accumulated. Besides, one month is a year's worth of time, right!? 1 second is 12 seconds. 5 minutes is 1 hour. 2 hours is 1 day. While you are reading a sentence like this, as every second passes in Africa, a second also passes in Japan.

Third: There is the talk that, on the flip side, people get turned off when they see AI-revealing phrases like "When there is bold text, I get turned off" or "That is extremely essential." Well, I do have a tendency to skip over things if I think, "Oh, they used generative AI to generate this," regardless of whether I actually use it or not, so this point resonates on an experiential level. That is why I include a note with high probability that "you can skip the trivial information that is everywhere." I don't personally mind bold text, so I think it's a matter of cultural spheres. Also, if I feel it is not interesting while reading it myself, I delete it or rephrase it.

In short, since I am taking up your valuable time, I don't intend to provide strange information, and I am writing based on the idea that it is fine if you can get some information out of it.

There is information that I could have expanded upon to release as an article, but I decided to pass on, which is as follows:

・autoresearch
・McKinsey's report on agents https://www.mckinsey.com/jp/~/media/mckinsey/locations/asia/japan/our insights/what is an ai agent/what is an ai agent___2k_0723.pdf
・The topic of what "GTM engineers" that global AI startups are desperate for are
・About 3 new OSS projects I'm building myself
・https://stripe.dev/blog/minions-stripes-one-shot-end-to-end-coding-agents
Explanation of the know-how in this article
・deer-flow
・The accuracy of Codex changes significantly depending on how well you organize operations—context, rules, validation, external connections, reuse, and automation—not just the cleverness of the model. What is the better way to do it?
・Talk about several models that have appeared recently (I am constantly watching whether they are higher or lower than Qwen3.5.)
・The paper "Experiences Build Characters: The Linguistic Origins and Functional Impact of LLM Personality"
・https://resources.anthropic.com/hubfs/The-Complete-Guide-to-Building-Skill-for-Claude.pdf
・MiroFish
・https://www.kali.org/blog/kali-llm-ollama-5ire/
・learn-claude-code
・openrag
・https://arxiv.org/abs/2603.04448
・https://arxiv.org/abs/2602.23643v1
・pua
・MiroThinker
・ATLAS V3
・AI tomato cultivation

Hey! That's way too long!!!!!! But looking at it, it seems like information from the last 2 or 3 days.
So, I am doing quite a lot of selection. This is how much information I am picking up.
What I want to say here is, if you are not someone who can discard information, not just output it, wouldn't you be swallowed by the waves of information?
It is true that there are too many release announcements lately, in particular.

↓ From here on, I have mostly let AI do the talking

Perhaps, the most important thing in content writing/generation by AI is not whether you used AI or not.
The important thing is who decided what to output and what not to output.

If you search, information will come up endlessly.
Even if you only follow the last 2 or 3 days, there are countless topics rolling around that could be turned into articles.
But if you pick up everything and output everything, the reader will be swallowed by the waves of information.

That is why I do not just stream what I pick up with AI.
I look at it myself, trim it, reorder it, and output only what I think will remain for some time after passing it through my own experience and judgment. Even if I say that, there are many stories where lasting about 3 months is sufficient, but that is still much better than streaming it without thinking at all.
Has the talk about OpenClaw or Mac mini already weathered away in the Japanese-speaking world?

I believe the essence of AI writing is not the speedup of text generation, but the redistribution of editorial responsibility.
The writer's job is not disappearing, but rather shifting toward "selecting," "discarding," and "taking responsibility."

This article is about my own mindset for that.
I don't know whether editing, praying, or primitive activity will be what remains.