AI for WordPress: setting up a magazine without vibecoding

Beyond Vibe Coding: Building a WordPress Magazine with Claude Code

For projects like an expert education journal, custom code isn't always necessary. Instead of diving into Next.js or Supabase, spinning up WordPress is often the smarter move: it comes with a ready-made admin dashboard, SEO plugins, a media library, and a REST API out of the box. The project author—an infrastructure specialist with zero PHP or WordPress experience—used the Claude Code CLI to provision a VPS and handle all the customization.

The goal: a minimalist blog featuring a hero section, an article grid, search functionality, and categorized sections. Development budget: $0. The result: a fully functional MVP built over a couple of evenings.

Infrastructure Setup

Stack: Ubuntu VPS, Nginx with caching, PHP 8.3 (FPM), MySQL 8.0, and the GeneratePress theme. Claude Code generated the SSH commands for:

• Installing packages and configuring PHP-FPM.
• Setting up Nginx with SSL (Let's Encrypt), a firewall (UFW), and Fail2ban.
• Deploying WordPress with pretty permalinks, plus Rank Math for sitemaps and SEO.

The workflow: define the task in plain English → AI writes the scripts → review and iterate. It saves hours on routine tasks, but manual verification remains non-negotiable.

Customization via mu-plugins

Twelve mu-plugins handled all the homepage logic. Sample prompts for the AI:

• A hero section featuring a pinned post, plus a three-column grid for recent articles.
• A dynamic category block using get_terms() instead of hardcoded HTML.
• Header search integration, sidebar removal, and hover effects on article cards.

Challenges and fixes:

• WordPress core styles overriding custom CSS → deployed a mu-plugin with priority 999.
• Grid breaking on mobile → targeted padding adjustments and refined media queries.
• Changing category slugs → switched from hardcoded values to dynamic queries.

Design choices: Inter font, white cards with subtle drop shadows, fully responsive layout. The AI generated the CSS based on plain-language prompts like "add a shadow on hover and lift the card slightly."

Iterative AI Workflow

A typical cycle looks like this:

• Define a precise task ("three-column grid, 20px padding").
• AI generates the PHP/CSS (roughly 80% accurate on the first pass).
• Test in the browser.
• Report the bug ("text touches the edge on mobile").
• Iterate on the fix (2–4 rounds per feature).

AI limitations to keep in mind:

• It can't see the rendered output—rely on detailed descriptions or screenshots.
• It doesn't track database changes—you must explicitly state dynamic requirements.
• Fixes can break adjacent code (file overwrites require careful version control).

Key Takeaways

• High ROI for non-experts: AI bridges gaps in the WordPress API (hooks, filters), saving weeks of documentation reading.
• Iteration is mandatory: Expect 2–4 cycles per task, but you won't need deep stack expertise.
• Baseline knowledge required: Familiarity with Nginx, SQL, and CSS is still essential for prompt engineering and validation.
• MVP-focused: Ideal for rapid prototyping and CMS customization (WordPress, Shopify).
• Security is manual: AI won't implement rate limiting or CSRF protection unless explicitly prompted.

When to Use This Approach

Best suited for:

• Prototypes and MVPs without strict production SLAs.
• Customizing familiar platforms on an unfamiliar tech stack.
• Solo developers or small teams.

Not recommended for:

• Mission-critical systems requiring enterprise-grade security.
• Users with absolutely zero technical background.
• Complex architectures lacking proper QA oversight.

The metrics: 2 weekends (3–4 hours each), VPS at ~$7/month, ~600 lines of CSS + 12 mu-plugins, plus a Claude subscription.

— Editorial Team