What started as a silly project became a proof of concept.
I almost listed it on Facebook Marketplace. A 2017 MacBook Pro, Core i7, 16GB RAM, 1TB storage. It’s been just sitting there, worth almost nothing on the secondhand market. I figured I’d get maybe $200 for it if I was lucky.
Instead, I turned it into a home server. I installed Ubuntu as an OS (so I don’t have to deal with ‘forever outdated’ MacOS nonsense on perfectly fine machine). And that decision quietly unraveled every assumption I had about how I build and host websites.
I wanted to share my journey as I reflect on AI as operational layer, self-hosted web stacks, and open-source software.
Some background
Before I started working at Automattic five years ago, I was a self-taught WordPress designer. I’d grab an all-in-one theme with a page builder from CodeCanyon and find a way to make it work. No deep technical knowledge, just determination and a lot of googling like “how to deactivate plugin without admin access” (Answer: FTP).
Five years of supporting WooCommerce merchants gave me a different kind of education. I got comfortable debugging without documentation. Trying to understand how product works without reading codes. Finding workarounds in products that systems weren’t capable of.
That skill… being comfortable in dark, unmapped infrastructure, turned out to be exactly what this story is about.
Phase 1: Own your AI infrastructure
I installed Ubuntu on the old MacBook Pro, partly to give it a second life, partly because an outdated macOS would have been the bottleneck anyway. Then set up Open WebUI (connected to Claude, ChatGPT via API, and local Ollama models) and exposed it to the internet through Cloudflare Tunnels .
The goal was simple: move my personal AI usage from scattered subscriptions to one coherent stack I actually owned.
One honest caveat here: this only makes financial sense if your actual usage stays below what a subscription would cost. And that math is different for everyone.
For me, ChatGPT was the easy call; I wasn’t using it in any meaningful way, so paying per API call instead of a flat monthly fee was obviously better. Claude was more complicated. API access works well for conversational use through Open WebUI. But if you’re running Claude Code heavily (which I now am) the subscription starts to look attractive again, because Claude Code’s usage can add up fast.
So “owning your AI infrastructure” doesn’t automatically mean ditching subscriptions entirely. It means being deliberate: knowing which tools you actually use, how you use them, and what each model of paying for them actually costs you. For me, this process was a really good practice to see the AI usage more clearly.
But finishing that setup surfaced a more interesting question: If I can own my AI stack, what else can I do?
Phase 2: Own your digital identity
That question led me straight to my email setup. This has been bothering me.
I’d been personally with Hostinger for years (for my work, I have access to WordPress.com and Pressable.com); affordable and decent, and they held everything: my domains, websites, about 20+ email inboxes bundled into my hosting plan. Convenient. (And yes, I hosted sites for free for friends and family…)
If I ever wanted to switch hosts, I’d be dragging 20 inboxes through a migration. Email migration is painful enough when you’re doing it once, let alone repeatedly. (It’s not that bad until you really have to do it. Don’t let me start on the email migration stuff.)
So I decoupled everything:
- All ~10 domains moved from Hostinger’s DNS to Cloudflare
- Email moved entirely to MXRoute — each domain now has hosting-independent inboxes
It took about five evenings. It was a whole ordeal. But the result: Now my stack is lighter and I don’t have to worry about email migration when I want to migrate to another host. Or I can spin up sites anywhere and just point Cloudflare DNS wherever I need.
Phase 3: Own your development infrastructure
Once I’d decoupled domains and email, rebuilding my sites was the obvious next step.
My portfolio had been neglected for years. It was still running on Elementor with a pile of free Elementor add-ons (and let me just say third-party built Elementor add-ons are eviliciously quick at degrading), and half the stack no longer worked.
I decided to use it as a test case for a workflow I’d been curious about: WP Studio (local WordPress environment) combined with Claude Code.
The combination worked well. WP Studio gave me a clean local environment Claude Code could actually read and reason about.
Here is my general prompting in phases:
- I first do “audit” phase where I let the Claude Code scan all the WordPress installation files of the old site. Then ask it to create “claude-audit.md” file of all the learnings it did.
- Once that is done, then I nail down my aesthetics I envision. I tell exactly what I want, and based on claude-audit.md file, I have Claude Code first generate claude.md file that has all the system prompt about the new site build.
With this two step, Claude Code has access to contents as well as plans for the outcome.
My Claude Code-activated WordPress stack is pretty simple:
- Build a custom theme with Tailwind. → No page builder bloat.
- No unnecessary plugins. Just a clean theme doing exactly what I needed. (You definitely need a few plugins still, and I will touch more about that later on.)
I first let Claude Code build everything hardcoded. Everything static. Once I see the design, then I instruct it to turn the page into Block based or to use Custom Post Type.
Then there it is, vibe-coded WordPress site that does not use page builder or any plugins.
My next big question was; where does this go? I didn’t feel like pushing it back to Hostinger. So I built a new infrastructure layer:
Total: roughly $160/year, plus Claude API usage. I built two sites in one night and spent about $8 in API costs, just FYI.
The whole workflow looks like this: install WordPress files using WP Studio, spin up a WordPress app in Coolify (~2 minutes), point Cloudflare DNS to the VPS (~5 minutes), and push whole WordPress installation folders to Coolify via SSH. A fully functioning site without a single unnecessary plugin.
And by the way, WP Studio allows to sync directly from WP Studio to WordPress.com or Pressable.com. And I’ve also built a site this way and synced. It worked perfectly. So instead, workflow can look like this:
- Spin up a local site via WP Studio
- Use Claude Code to develop
- Deploy directly to WordPress.com or Pressable.com hosted site
I have to be honest: I work at Automattic, so you might think I am trying to sell you Automattic-built products. But this workflow was an easy winner for me because it works so nicely. If I could test, I would test my workflow with Local WP or DevKinsta too. You get my point.
I’ve also tested the following workflow:
- Pull WordPress site files via SSH to local files
- Use Docker to set up ‘local dev site’ that’s live during editing via Claude Code
- Whenever changes are made and ready, you can then push the confirmed changes to the server
Claude Code-assisted WordPress development works well with any host that gives you real, unrestricted SSH access, which is standard on VPS hosting and most developer-oriented hosts, but not guaranteed on shared or fully managed plans. So I don’t know how that would truly look like with a normal managed hosting.
What’s actually novel here
I want to be precise about this, because I don’t want to oversell this as ‘revolutionary.’ It’s a proof of concept.
None of the individual pieces are new either: VPS hosting, self-hosted WordPress, open-source panels like Coolify — these have existed for years. What changed is the combination:
- An AI operator (Claude Code) that can SSH into a server, inspect containers, understand code at project scale
- Structured context — audit files that let Claude understand a site as a system, not just as isolated files
- Infrastructure you actually control — so the AI can participate in production safely
- Speed that makes it practical — multiple sites rebuilt before lunch
Together, a solo builder could operate at a pace that previously required a team. (Well, frankly speaking, this was all impossible for me with the level of technical knowledge I have.) AI can now be a full participant in the development workflow — not just an assistant at the edges.
What this means for the WordPress ecosystem
None of the individual pieces here are new. VPS hosting, self-hosted WordPress, and open-source panels have existed for years. What changed is the combination:
- An AI operator that can SSH into a server, inspect containers, and understand code at a project scale.
- Structured context that lets AI understand a site as a complete system, not just isolated files.
- Speed that makes custom development faster than visual clicking.
A solo builder with basic systems intuition can now operate at a pace that previously required an agency team. In retrospect, what I stumbled into has massive implications for the industry.
What Gets Shaky
- Page Builders: They exist because non-technical users (like me) needed a way to customize sites without writing code. Page builders optimized for visual decision-making (click, drag, render). But AI codes faster than you can click. If I can describe what I want, get the code, and iterate in 30 seconds, the math no longer favors the page builder.
- ACF: You can just build Custom Post Type or Custom meta fields directly with no technical knowledge.
- Utility Plugins: If a plugin relies on 100 lines of code or less, Claude Code can build it directly into your theme in two minutes. The market for “does one small thing” plugins will contract significantly.
What Thrives
- WordPress Core: WordPress is a CMS, and that doesn’t change. What changes is how it gets built. Custom theme development used to be slow and expensive; now it is fast and cheap. WordPress gets simpler, more flexible, and more essential.
- WooCommerce: Commerce logic is genuinely complex—inventory, payments, taxes, refunds. AI can build the storefront, but it can’t architect the business logic. WooCommerce becomes the system you automate the building of, not the system you replace.
- Any service layers such as: SEO, Ads, Marketing integrations, Security, Caching/Optimization, MAM …
- Infrastructure Layers: Cloudflare is becoming the default outermost layer of the web. Cheap VPS providers and control panels like Coolify will surge in value as developers seek performance without platform lock-in.
The real change is coming
When implementation becomes more affordable, competitive advantage shifts entirely to judgment.
What should the site actually do? What content matters? What is the brand direction? The person who can direct the system effectively becomes vastly more valuable than the person who just operates the tools.
The common narrative is that AI makes web development easier. My version, after living through this, is that AI has become part of the operating system—entirely changing how WordPress sites will be built going forward.
I have a fine arts degree. I’ve never called myself a developer. But I am good at understanding systems, debugging, directing AI, and making judgment calls about design. That combination—system intuition plus AI fluency—turned out to be enough to own and operate a complete web infrastructure stack.
The barrier to this specific way of working just collapsed. And it’s only one prompt away from being unlocked for anyone willing to ask.