One Person Maintains 30+ WordPress Plugins with AI: How Generative Tools Are Changing Plugin Strategy

– How generative AI can dramatically improve plugin maintenance — from code review and documentation to testing and dependency updates.
– How AI assistance changes the way developers think about ownership and sustainability of open-source projects.
– Practical workflows for integrating AI tools into daily plugin development without losing human creativity.
– How this shift impacts WordPress agencies: moving from reactive maintenance toward proactive, scalable innovation.

Transcript

1. A 1 Person Maintains 30+ WordPress Plugins with AI: How Generative Tools Are Changing Plugin Strategy Fumiki Takahashi @WordCamp Asia 2026
2. Who is Fumiki Takahashi • CTO at TAROSKY Inc. • 20 years in the WordPress community • Lead Organizer, WordCamp Tokyo 2016 • SF fiction writer & Founder of Hametuha 2
3. 30+ plugins on WordPress.org 50+ repositories on GitHub 3 Maintained by... 1 person.
4. You may think: “Useless. That’s reinventing wheels. Use exiting plugins.” But: • Yahoo! News FTP submission — ~22.5B PV/month • Sitemap for 200,000+ posts under bot traffic • Rankings via Google Analytics, not WordPress DB Why So Many? 4
5. The Real Problem • Shipping is easy. Maintaining 30+ repos is not. • Stale repositories pile up silently • Big waves hit all at once: Block Editor, CI Migration, Building tools. • Fixing 30+ repos at once. Hard work 󰷹 • No time for documentation. 5
6. Demographics 6 Souce: prb.org 2024 Turkey 1.5 India 2.0 Iran 1.7 Thailand 1.0 Indonesia 2.2 Philippines 1.9 2.1 = Replacement level Following Countries
7. What’s happening in Japan in 2026? • 29%+ of population is 65 or older • Losing 1,000,000 people per year • Younger talent is less and more expensive • Web Development is less attractive 7
8. "Take only simple projects. " 8
9. The Turning Point 9
10. AI Agents(2025) 📝 Test Cline: "Repository A is modern. Repository B is old. Make B look like A." 👉 ✅ Done. "Maybe I can actually handle 30 plugins." 10
11. Copilot vs. Agent 11 Copilot Agent Who operates You AI Your role Driver Director Unit of work Line / Function Task / Repository Scalability 1 person = 1 stream 1 person = N stream
12. CLI: • Claude Code（😘） • Codex CLI（by Open AI） • Gemini CLI（by Google） IDE integration/extension: • Cline • Cursor • Antigravity（by Google） Agentic Coding Tools 12
14. • Don't get attached to a specific model. Models’ superiority change frequently. • Evaluate by: Cost / Data policy / Political context • Focus on the “harness engineering” features. On model choice 14
15. 15 By Internet Archive Book Images - https://www.flickr.com/photos/internetarchivebookimages/14597244417/Source book page: https://archive.org/stream/turfgoodsfinehar00gsel/turfgoodsfinehar00gsel#page/n45/mode/1up, No restrictions, https://commons.wikimedia.org/w/index.php?curid=44233023
16. How AI Agents leverage plugin development 16
17. Strategy • Learn from great repositories • Standardize repo structures ◦ Add Evaluation Tools ◦ Agent Skills • Apply to all repositories at once • Add AI review and triage • AI-powered suggestions for improvement • Make feedback loop 17
18. Learn from Great Repositories Prompt to Agent: "Compare my-plugin rich-taxonomy with site-kit-wp. Identify best practices we should adopt — especially around testing and CI/CD." ✅ Done. “Summarize todos in TODO.md” 18 works: 3m 39s
19. Standardize Repository Structure Common rules applied across all 30+ plugins: • README.md → auto-converted to readme.txt by wp-readme • PR required → merge → auto-draft Release → publish → deploy • JS transpile/lint: unified with wp-scripts • Business logic: includes/*.php • Large projects: PSR-0/PSR-4 under app/vendor/name • Build: src/{js, blocks, scss} → dist/{js, blocks, css} The standardization work itself? Done by the Agent. 19
20. Add Evaluation Tools Context is words(tokens) = 🔋 battery for Smartphone Wasted context = lower accuracy + hallucination Let tools answer "is the code correct?" So the Agent doesn't have to think deeply. • phpcs — WordPress coding standards • phpunit — Test suite for PHP • phpstan + phpstan-wordpress — PHP static analysis • eslint / stylelint — JS and CSS(Use @wordpress/scripts) • husky — auto-lint/fix before commit • playwright — E2E test tool(efficient, but heavy 󰷹) 20
21. Use Agent Skills Agent Skills = reusable instruction sets for the Agent(prompts & codes) WordPress official skills: → github.com/WordPress/agent-skills TAROSKY custom skills: → github.com/tarosky/claude • glotpress-translate — Download translations, AI-fill untranslated strings, flag for human review • wp-ci-setup — Interactively set up GitHub Actions from the best-performing plugin; auto-switches for public vs. private • t-wada — Adds missing tests based on TDD philosophy by t-wada (pioneer of TDD in Japan) 21
22. Run Agent Across Multiple Repositories Key concept: • Place multiple repos under a parent directory • Run the Agent from the parent level • Agent operates on all repos in parallel Anthropic Released “Agent Teams” 22
23. Demo Video 23
24. 24 8x faster
25. Review Automation On GitHub, Install Claude app and setup GitHub Actions e.g. taro-open-hour#45 Guardrails: • Avoid cost abuse • GitHub default protection 25 claude-review.yml
26. What WordPress' official AI guidelines say “Can AI be the sole reviewer for my code?” 👉 NO😣 26 CODE REVIEW OUTPUT CODE REVIEW OUTPUT Before AI After AI New Bottleneck REMINDER: Japan loses 1M people per year.
27. Improvement 27 Keeps AI working on: • Proposing new issues • Triage issues • Researching competitors • Check forum posts Plan Do Adjust Check
28. 28 https://github.com/tarosky/rich-taxonomy/issues
29. Three Things to Do This Week 1. If you haven't used a Coding Agent — start today. 2. Let AI do what you've been skipping. 3. Don't give up on your stale repositories. 29
30. Thank You! Catch Me and ask anything. 30