Hi, I'm Shingo Suzuki, also known as Suzushin.
Recently, I substantially reorganized my blog's editorial team—the collection of Claude Code skills I use. The biggest change was deciding to write the article text myself. Here's why I changed direction and what the new team looks like.
I've substantially rebuilt the agent team
In my earlier article about creating a Claude Code agent team for this blog, I described seven roles: secretary, brainstorming, SEO, interviewer, writer, proofreader, and marketer. Using them in practice revealed some problems.
For this redesign, I reconsidered the roles and organized the skills that had been added or changed.
Why I stopped having a writing agent draft articles
Previously, the writing agent imitated my style and wrote articles. I was reasonably satisfied with the quality. Still, something often felt off—not only in isolated passages, but gradually throughout the whole article.
What bothered me most was whether I felt comfortable with the content itself. AI-written text didn't feel fully like my own words, and rereading it often felt strange. “Can I really call this my blog article?” I sometimes wondered. Since it's my own blog, I increasingly wanted to write in my own words.
So I changed from having AI write for me to writing together with AI.
Claude Code now supports outlines, substantive feedback, and proofreading to help improve quality. Writing the body myself lets me feel more satisfied that the words are mine.
Meet the new editorial team
These are the current article-related skills:
- note: secretary
- brainstorm: ideas
- seo-brief: SEO
- interview: interviews
- outline: outlining, renamed and repurposed from the writing agent
- feedback: substantive review, newly added
- proofread: proofreading
- monetize: monetization
- note-write: drafts for note
Previously, the writing agent ghostwrote the body. I've removed that role and renamed it outline, specializing in structure instead.
Another important change is separating feedback, for content review, from proofread, for mechanical checks. Feedback reviews substance, logic, and appropriate expression. Proofread checks things such as typos and grammar.
Overall, I now write the body myself and use AI as support.
How it feels in use—this article is an example
This article itself is being made with the new setup. Brainstorm supplies ideas, interview draws out firsthand experiences, outline creates a structure, feedback reviews the substance, and proofread checks the text. I write the article body.
Writing everything myself takes effort, but an outline makes it easier than facing a completely blank page. Content review helps me check the logic and wording, so I feel happier with the quality too.
Using AI from brainstorming through outlining, review, and proofreading has let me improve the article while still writing in my own words. That feels like a major gain. The process went relatively smoothly, so I'd like to keep using it.
What I'd like to try next
I'll continue with this arrangement for a while. To improve articles and outlines further, I'd like to work on:
- Freshness checks: detecting old information and whether updates are needed
- Prioritization: deciding which articles to write first and which can wait
- More accurate substantive review with concrete improvement suggestions
The phrasing may differ a little from earlier articles. But this is closer to how I naturally write, and working with AI seems to help me produce something more recognizably mine.
This new approach has only just begun. I hope you'll look forward to seeing how it develops!