This blog had exactly two posts.

“Starting My Blog” and “Portfolio.” That’s it. I started in January, and it was already February. Not great for a blog that’s supposed to be my portfolio and growth diary.

It’s not that I didn’t want to write. I just kept getting stuck at “what should I write about?” Between teaching high school full-time, lesson planning, and building side projects, I never had the mental bandwidth to brainstorm blog ideas, so nothing got written.

Then on February 5th, Anthropic released Claude Opus 4.6.

The headline feature was Agent Teams — a system where multiple AI agents work as a coordinated team, each with a specialized role, running in parallel instead of one agent doing everything sequentially. Claude Code got this agent mode as well.

The moment I saw the news, my motivation to blog came rushing back.

“What if I built a blog management team with this?”

The Spark: Inspired by Opus 4.6’s Agent Teams

I’d already been using Claude Code heavily for this blog. It’s built with Astro and deployed on Cloudflare Pages, and Claude Code helped with everything from drafting content to technical maintenance.

But asking for help on a case-by-case basis meant each interaction was isolated. With Agent Teams now available in Opus 4.6, I could do something more systematic — build an organized operation to keep the blog running consistently.

Claude Code has a feature called custom subagents: by placing Markdown files in .claude/agents/, you can define specialized agents with specific roles and have them work together as a team.

So I asked Claude Code directly:

“I want to manage my blog with Claude Code. Can you design an organizational structure? Like, with a project manager?”

Claude Code’s First Proposal: 6 Custom Agents

Claude Code came back with a 6-agent structure:

A clean separation of concerns. I was impressed.

But something was nagging me.

The Real Bottleneck: “What to Write”

A Writer can’t write without a topic. A Translator can’t translate without a source article. The whole pipeline was useless if nothing was feeding into it.

My blog’s problem wasn’t writing ability — it was the lack of a system to turn ideas into articles.

I had plenty of material from my daily work and learning. But the jump from “that’s interesting” to “let me structure this into a blog post” was too big.

When I explained this to Claude Code, it gave me a surprising response.

“You already have a goldmine of content.”

It scanned all my existing projects and listed over 12 potential article topics:

Every single one was something I’d already built. I didn’t lack ideas — I lacked someone to extract them.

Agent #7: Content Strategist

That’s when we added the Content Strategist.

Not just an “idea person” — a strategist. The job is clear:

1. Scan existing projects for article-worthy topics
2. Prioritize by SEO value, writing effort, and portfolio impact
3. Create outlines detailed enough for the Writer to start immediately

That third point is key. “Hey, here’s an idea” alone leads to “Cool, I’ll write it someday.” But when someone hands you a structured outline with headings and key points for each section, the barrier to actually writing drops dramatically.

Agent #8: Chronicle (This Is the Fun One)

After finalizing the 7-agent design, something clicked.

“This design process itself is a blog post.”

The conversation I was having with Claude Code — discussing problems, weighing options, arriving at decisions — that is the content for my blog. A teacher-developer building a blog management team with AI. That’s a story worth telling.

But these “processes” get forgotten if you don’t write them down.

So we added an 8th agent: Chronicle.

Chronicle’s job is to record the collaborative process with Claude Code itself. Design discussions, technical decisions, failures and fixes — all structured and stored so the Content Strategist can turn them into article ideas.

In other words, running the blog generates content for the blog. A self-sustaining loop.

The Final 8-Agent Structure

Blog PM (Project Manager)
 ├── Content Strategist (Ideas + Outlines)
 ├── Writer (Japanese articles)
 ├── Translator (English versions)
 ├── Reviewer (Quality checks)
 ├── SEO Analyst (Search optimization)
 ├── Tech Maintainer (Technical upkeep)
 └── Chronicle (Meta-documentation)

The standard article workflow:

Content Strategist proposes topic + outline
  → Writer drafts Japanese article
    → Reviewer proofreads Japanese
      → Translator creates English version
        → Reviewer proofreads English
          → SEO Analyst optimizes metadata
            → Blog PM confirms build → Publish

Since quality is the priority, the Reviewer checks both the Japanese and English versions separately.

Implementation with .claude/agents/

Each custom subagent is defined as a Markdown file in the .claude/agents/ directory. These files contain the agent’s role, guidelines, and constraints.

myblog/.claude/agents/
├── blog-pm.md
├── content-strategist.md
├── writer.md
├── translator.md
├── reviewer.md
├── seo-analyst.md
├── tech-maintainer.md
└── chronicle.md

For example, here’s a snippet of the Writer agent definition:

# Writer

## Role
- Write Japanese articles based on outlines from Content Strategist
- Use first-person "僕" (boku) — one of several Japanese first-person
  pronouns, this one conveys a casual, approachable tone
- Match the style of existing articles

## Style Guidelines
- Casual but not sloppy, technical but accessible
- Write from personal experience ("here's what I tried", "here's what happened")

By defining each agent’s role and constraints in Markdown, you get consistent behavior every time you invoke them. The Translator has instructions to produce natural English rather than literal translations. Each agent knows its place in the workflow.

The project’s CLAUDE.md was also updated with the full team structure, so any agent can understand the overall organization.

Reflections on AI-Powered Blog Management

Honestly, I don’t know yet if this system will work at scale. That’ll become clear once I actually start producing articles with it.

But the design process taught me something important.

The content was already inside me. What was missing was a system to extract it, structure it, and get it ready to write.

When the Content Strategist scanned my projects and pulled out 12+ article ideas, I was genuinely surprised. I’d built all of those things, but it never occurred to me that they could be blog posts.

And this very article? It was born from Chronicle’s first log entry — a record of the agent design process. The meta-documentation system worked on day one.

What’s Next

• Test the 8-agent workflow with real articles
• Have the Content Strategist build a prioritized topic backlog
• Iterate on the workflow based on what works and what doesn’t
• Keep feeding Chronicle with every meaningful process

This blog started with 2 posts. It’s about to change.

See ya!