2026.03 · Pebblous Data Communication Team
Reading time: ~20 min · 한국어
Executive Summary
The Pebblous blog, which began with its first commit in September 2025, achieved 57 articles and 753 commits within 3 months. Three months later, the blog has grown to 79 published articles, 1,013 commits, and 21 project directories. This is not merely a numerical increase — the very way the blog is built has fundamentally changed.
The biggest transformation is the shift from "vibe coding" to "agentic automation." Six Claude Skills automate the entire process from article creation to SEO verification, RSS/sitemap publishing, and SNS copywriting, while 4 GitHub Actions compose the CI/CD pipeline. The PebblousPage.init() module system initializes 10+ components with a single config object.
This report dissects the current state of the Pebblous blog. It documents what articles exist, how they are created, and what automation systems are in operation — archiving the birth of an agentic blog.
1. Key Metrics
Six months since the first commit on September 28, 2025. Here is the Pebblous blog by the numbers.
Monthly Article Publication Trends
2. Growth Story: From 57 to 79
In the 2025 Year-End Report, we documented a "3-month sprint." 57 articles, 753 commits, 86,000 lines of code. It was the first milestone achieved through vibe coding and AI-assisted writing.
- Published Articles57
- Commits753
- Projects12
- Bilingual0 pairs
- AutomationManual
- Dev ApproachVibe Coding
- Published Articles79 (+38%)
- Commits1,013 (+35%)
- Projects21 (+75%)
- Bilingual7 pairs
- AutomationSkills 6 + Actions 4
- Dev ApproachAgentic Automation
Over 3 months, 22 new articles were added and 9 new project directories were created. Notably, new research areas such as NeuroSymbolicAI, World Model, BizReport, SyntheticData, AgenticAI, and PebbloSim appeared on the blog.
Article Distribution by Category
But more important than these numbers is the change in how the blog is built. In December 2025, it was "vibe coding" — writing markdown and chatting with Claude to generate HTML. By March 2026, an agentic automation system is in place where a single /new-post command kicks off article creation, and /publish automatically regenerates the RSS feed and sitemap.
3. Agentic Blog Architecture
An "agentic blog" is a blog where an AI agent (Claude Code) can autonomously perform the entire process — from article creation, SEO optimization, and publishing to promotion. Humans provide the topic and manuscript; the agent handles the rest.
3.1 Claude Skills (6)
Claude Skills are reusable AI agent task units defined in the .claude/skills/ directory. Each skill has its workflow specified in a SKILL.md file, enabling Claude Code to produce outputs of consistent quality.
3.2 GitHub Actions (4)
A CI/CD pipeline that automatically runs on git push. Without human intervention, the sitemap, RSS, and search engine indexing are automatically updated.
3.3 PebblousPage.init() Module System
Every article page is initialized with a single config object. 10+ modules are automatically loaded, handling everything from header, footer, and theme to schema, related posts, and comments.
| Module | Purpose |
|---|---|
| PebblousTheme | 3-theme system (dark/light/beige) |
| PebblousPage | Page initialization, dynamic Hero section rendering |
| PebblousChart | Chart.js wrapper (bar, doughnut, pie, line) |
| PebblousTabs | Tab component |
| PebblousComponents | Collapsible, Accordion |
| PebblousUI | Progress bar, Back-to-top |
| PebblousComments | giscus comment integration |
| PebblousRelatedPosts | Related post recommendations based on articles.json |
| PebblousBreadcrumbs | Breadcrumb navigation |
| PebblousSchema | Auto-injection of JSON-LD schema (FAQ, Article, Breadcrumb) |
Core Principle: The <h1 id="page-h1-title"> in the Hero section must be left empty.
PebblousPage.applyConfig() dynamically fills it from config.title. FAQs are managed exclusively via config.faqs — never through HTML microdata or <head> JSON-LD.
4. Content Pipeline
This is the complete process from a single markdown manuscript to a published blog article. At each step, Claude Skills or GitHub Actions operate automatically.
Chain automation triggered by a single JSON:
articles.json is the Single Source of Truth for this blog.
Article titles, paths, categories, dates, and keywords all live here, and RSS generation, sitemap updates, index page rendering, and related post recommendations all derive from this single JSON file.
When a new article is registered in articles.json and pushed, GitHub Actions automatically regenerate sitemap.xml and rss.xml, ping Google, and notify search engines via IndexNow.
Optionally, you can add bilingual conversion with /bilingual and SNS promotional writing with /sns-write. Throughout the entire process, humans decide only the "what" — the agent handles the "how."
5. Blog Inventory
79 published articles are distributed across 4 categories and 21 project directories.
Project Directory Map (21)
Bilingual Articles (7 Pairs)
Seven pairs of Korean/English articles converted using the /bilingual skill. hreflang tags ensure language-specific exposure is separated in Google search results.
Full Article List
Dynamically loaded from articles.json. Organized by category.
Tech
Data Art Lab
Business
Story
6. Tech Stack Evolution
Here is how the tech stack has expanded compared to the 2025 year-end report. Newly added elements are marked with NEW.
Frontend
- HTML5 + Tailwind CSS (build)
- Vanilla JavaScript (no framework)
- Chart.js (data visualization)
- 3-theme system (dark/light/beige)
- NEW PebblousPage module system (10+)
- NEW giscus comment integration
Backend · CI/CD
- GitHub Pages (static hosting)
- GitHub Actions (CI/CD × 4)
- Node.js (Tailwind, Sitemap, RSS)
- NEW IndexNow protocol
- NEW OG image auto-generator
Development Tools
- Visual Studio Code
- Claude Code (agent)
- NEW Claude Skills (×6)
- NEW CLAUDE.md (project guidelines)
- NEW 17 internal docs (docs/)
Key Change: In 2025, AI was an "assistant that writes code." In 2026, a "project rule system that AI understands" has been established. CLAUDE.md, 17 internal documents, and 6 Skills combine so that the AI agent understands project conventions and produces outputs of consistent quality. This is the difference between vibe coding and agentic automation.
7. 2026 Outlook
In the 2025 year-end report, we anticipated "publishing more than 200 articles." Now that the agentic automation system is in place, that goal is more realistic than ever.
Content Expansion
- 200 articles goal — agentic pipeline enables 15-20 publications per month
- Bilingual content expansion — expanding English versions of key articles for global reach
- Data Art Lab expansion — beyond code painting to interactive data art
Technical Advancement
- Deeper agentic automation — AI performs topic suggestions and trend analysis
- Automated SEO monitoring — auto-improvement based on Google Search Console data
- Community building — expanding reader engagement through giscus comments and newsletter integration
"A blog that leverages AI while retaining a distinctly human touch" — this pledge from the 2025 year-end report remains in full effect. Agentic automation is merely a tool; the direction and values of the blog are determined by people. The Pebblous blog will continue to deliver deep insights on AI-Ready Data, Physical AI, and Data Quality — leveraging technology to the fullest for faster, more prolific, and more consistently high-quality output.
8. Future Works
The infrastructure and automation systems have matured. The next leap depends on strategic content balance, enhanced art visibility, and data-driven decision-making. The intersection of art, tech, and business — this is the moat that no other corporate blog can replicate, unique to Pebblous.
8-1. Rebalancing Content
Currently, the Tech category accounts for 49% of all content, while Story (8%) and Business (19%) are relatively weak. To realize the art-tech-business convergence the brand promises, the business narrative must be strengthened. Strategic content such as customer insights, market positioning, and "why data matters to the CEO" is the answer.
8-2. Rediscovering Art
There are 25 data art pieces, but the Featured rate is only 12% (compared to 30% for Tech). Art content is a unique asset that no other corporate blog possesses. A gallery-style art showcase landing page should be created so this differentiating factor can speak for itself.
Action Items: Expand Featured art pieces to 6-8, create a new art showcase page with technique/theme filtering, extend Data Art Lab beyond code painting to interactive data art
8-3. Data-Driven Content Decisions
Content production automation is at a top level, but there is no visibility into what resonates. A /content-analytics skill that maps Google Analytics data to articles.json categories could close this loop. Which topics drive traffic, which language versions perform better — right now, there is no visibility.
Required New Skills
- /content-analytics — integrated analysis of GA data + articles.json
- /content-calendar — quarterly theme planning, category gap analysis
Measurement Metrics
- Traffic and dwell time by category
- Performance comparison by language version (KO/EN)
- Chart interaction and scroll depth tracking
8-4. Cross-Pillar Content — True Differentiation
Not art separately, tech separately, business separately — the magic of Pebblous lies at the intersection of these pillars. Content like "Visualizing ISO compliance data as generative art" or "The business case for data aesthetics" is something only Pebblous can create. This intersection is a unique territory that no competitor can replicate.
8-5. Bilingual Completion
Currently, 57% of posts are available in KO+EN bilingual format. Resolving the remaining 43% (19 articles) of single-language debt would significantly expand global reach. The /bilingual skill is already in place, so prioritizing high-performing articles for conversion maximizes the return on effort.
Priority Roadmap
P1: Immediate (within 2 weeks)
- - Expand Featured art pieces to 6-8
- - Establish content calendar (quarterly category targets)
- - Normalize changelog system usage
P2: Short-term (within 4 weeks)
- - Develop /content-analytics skill
- - Convert top 10 single-language articles to bilingual
- - Batch publish 4 business content pieces
P3: Mid-term (within 8 weeks)
- - Create art showcase landing page
- - International SEO audit (hreflang, per-language traffic)
- - Plan cross-pillar content series
P4: Long-term (Q2 2026+)
- - Explainer video content (tech topics)
- - Medium, Dev.to content syndication
- - Community forum/discussion channel expansion