vibeyour website Hire us
DIY Vibe Coding

Vibe Coding Case Study: Three Real Sites, Built by One Crew

August 16, 2026 · 9 min read · By Vibe Your Website Agency Team

Every vibe coding case study you’ve read was written by someone selling a course. This one is written by developers showing three websites they actually built, with the boring parts left in. The first site is the one you’re reading right now. The second is Sminkopedia, a Serbian-language beauty encyclopedia with 219 published articles that our crew built and runs. The third is HR Blizz, a marketing site for enterprise payroll software that serves 160-plus countries, which is about as far from a beauty blog as web work gets. Same team, same method, three completely different businesses. Here’s what the AI did, what the humans did, what broke, and what we’d do differently, so you can decide whether this approach fits whatever you’re trying to build.

Case study 1: VibeYourWebsite.com, the site that sells the method

Yes, we vibe coded the website that sells vibe coding. It would be a strange advertisement if we hadn’t.

The brief we gave ourselves: a site that does three jobs at once. Give away a real prompt library free, publish flat prices for people who’d rather hire us, and rank for the questions people actually type about vibe coding. No page builder, no bloated theme from a marketplace, nothing we’d be embarrassed to hand a client.

What got built: a custom WordPress theme, written from scratch. Hand-rolled CSS design system with design tokens, no Tailwind, no build step, so any developer can open a file and read it. A prompt library of 40 free prompts (around 12,000 characters each) organized by platform and by 40 business types. Nine service pages with prices printed on them. Sixty-plus blog posts. A schema layer that marks up FAQs, services, and offers so search engines and AI assistants can quote us accurately. Security hardening that shuts off the usual WordPress attack surface. An llms.txt file that hands AI crawlers our sitemap and prices directly.

What the AI actually did: the first draft of almost everything. Template scaffolding, the CSS token system’s first pass, section layouts, schema JSON structure, even the security header list. On a custom theme like this, AI comfortably wrote the majority of the initial code. That part of the hype is real, and it’s why the whole build took weeks instead of the months a from-scratch theme used to take us.

What the humans did: everything that makes it production instead of a demo. Deciding that author archives should redirect because they leak login names. Catching that the AI’s first drafts confidently recommended WordPress plugins that do not exist (it does this a lot; check every plugin name it gives you). Making forms actually deliver mail. Keyword research against what people really type, then rewriting titles the AI had guessed at. Moving the debug log out of the public webroot, where a surprising number of live WordPress sites keep theirs.

What broke along the way: plenty. Early favicon setup served an SVG that Google ignores (Google wants PNG or ICO, at 48 pixels or bigger). A WP-CLI command that silently failed on large files cost us an evening. None of it was exotic. All of it was the kind of thing you only catch if someone on the project has been shipping websites for years.

What a build in this class costs: this is our standard WordPress tier, $5,800 flat, 7 to 10 days, code in your GitHub.

Case study 2: Sminkopedia.com, a content business in another language

Sminkopedia is a beauty encyclopedia for the Serbian market: makeup tutorials, product reviews for what’s actually sold in Serbian drugstores, coverage of regional beauty events from Novi Sad to Istanbul, and a city guide for finding makeup in Belgrade. It currently has 219 published articles and it grows every week.

Why it matters as a case study: it’s the opposite of this site in almost every way. Different language, different audience (beauty readers, not developers), and a business model built on content volume instead of leads. If vibe coding only worked for tech-adjacent sites in English, Sminkopedia is where it would have fallen apart. It didn’t.

What got built: a custom WordPress theme again, because the whole point of a content site is that editors publish without a developer in the loop. Category architecture that mirrors how readers browse (tutorials, looks, trends, events), templates that make a 2,000-word tutorial readable on a phone, and an events section for fair announcements that would be miserable to maintain by hand.

What the AI did: the theme scaffolding and layout work, same as before, plus something less obvious: it made building in Serbian a non-issue. Vibe coding tools don’t care what language your content is in. The prompts were English, the output structure was universal, and the editorial layer stayed fully Serbian.

What the humans did: editorial standards, real product knowledge, and the publishing discipline that 219 articles requires. No AI writes a trustworthy review of a mascara sold in Belgrade. The tech stack just makes sure nothing stands between the person who knows and the reader.

The lesson we’d underline: for content businesses, the build is maybe 10% of the work, so the correct move is to make the build cheap and fast, then spend everything you saved on the content itself. Vibe coding is precisely that trade.

What a build in this class costs: the same $5,800 WordPress flat fee covers a content site like this; ongoing publishing is the owner’s investment, not ours to meter.

Case study 3: HR Blizz, the enterprise B2B site

The third case study kills the last comfortable excuse, the one that goes “sure, vibe coding works for blogs and small shops, but serious companies need serious agencies.” HR Blizz is the marketing site for a global payroll platform that processes payroll in more than 160 countries. The audience is enterprise HR and finance leadership. The sales cycle is long, the compliance stakes are real, and the site has to hold its own against competitors with eight-figure marketing budgets.

The brief: explain a genuinely complicated product (gross-to-net payroll, approval workflows, validation engines) to buyers who have seen a thousand SaaS sites, without drowning them in the usual enterprise word soup. The centerpiece is an interactive console preview right on the homepage: instead of screenshots, visitors step through the product’s actual workflow surfaces, leave requests, employment certificates, validation rules, before ever talking to sales.

What got built: once again a custom WordPress theme, built for this one product. Interactive demo sections that behave like the software they’re selling. Page architecture shaped around how enterprise buyers actually evaluate: capabilities, countries, compliance, then proof. All of it fast, because a slow site selling payroll automation is a joke that writes itself.

What the AI did and didn’t do: same split as the other two builds, just at higher stakes. AI generated the bulk of the interactive component scaffolding and layout code, which would have been weeks of hand work in the pre-AI era. Humans decided what the demo should show, what enterprise buyers need to see in the first ninety seconds, and how to write about payroll without inducing sleep. The judgment layer doesn’t shrink as projects get bigger. It grows.

The lesson: the method scales up. A B2B site with an interactive product demo and enterprise credibility requirements came out of the same workflow as a beauty encyclopedia. What changed was the number of human decisions, not the fundamental economics.

What all three builds have in common

Strip away the differences and the same skeleton shows up in every project:

  • Custom theme, no page builder. AI writes clean code fast enough that builder bloat is no longer a trade worth making.
  • WordPress where non-developers publish. Both sites are edited by people who never touch code. That constraint drives more architecture decisions than any technology preference.
  • AI for the first 80%, experience for the last 20%. The draft is cheap now. Judgment about security, SEO, and edge cases is where the value moved.
  • Search-first structure. Both sites were shaped around what people type into a search box, not what looked nice in a mockup.
  • Boring, durable hosting and backups. Nobody’s proud of this part. It’s why neither site has had a data-loss incident.

What these case studies say about your project

If your project resembles any of these, a lead-generating business site, a content site with real publishing ambition, or a product site that has to convince serious buyers, the method transfers directly. The honest boundary line for when vibe coding stops being appropriate, and the step-by-step version of the method, are both in how to vibe code a website.

If you want to attempt it yourself, start with how to vibe code a website and the free prompt library. Those prompts are the same starting points we used on both of these builds. The realistic budget for DIY is about $20 a month while you build, and the realistic cost breakdown is in what a vibe coded website costs.

And if you’d rather have the crew from these two case studies build yours: flat prices are public, $2,800 for static sites, $5,800 for WordPress, and the form takes two minutes.

Hire the crew behind these case studies

Every case study above is a sales pitch, so here is the honest version of the close. We are the vibe coding agency that built all three sites, and the vibe coding pricing we charge is printed publicly: $2,800 static, $5,800 WordPress, stores from $7,800, small tasks at $60 an hour. You get a real answer within a day, nightly progress videos, a deadline we hit, and the repo in your GitHub from day one. If the three builds on this page look like the kind of work you want on your own project, the form takes two minutes. If you want to check our numbers against anyone else first, good, here is how to compare us.

Vibe coding case study FAQ

Is a vibe coded website good enough for a real business?

All three sites in this case study are real businesses. One generates client leads in the US market, one is a growing 219-article content property, and one markets enterprise payroll software to global buyers. The qualifier that matters isn’t whether AI wrote the first draft of the code, it’s whether someone competent handled security, SEO, forms, and hosting before launch.

How long did these vibe coded builds take?

Weeks, not months, for the initial build in each case. The content is a separate, ongoing effort: a prompt library and 60-plus posts on one site, 219 articles and counting on another, and continuously updated product pages on the third. Vibe coding compressed the code timeline dramatically and did nothing to compress the writing, which is exactly what you should expect.

Does vibe coding work in languages other than English?

Yes. Sminkopedia is entirely in Serbian. You prompt the AI in whatever language you like (English tends to get the best code output), and your site’s content language is a completely separate decision. The tooling is language-agnostic about what you publish.

What went wrong in these builds?

AI-invented WordPress plugins that don’t exist, a favicon format Google ignores, a debug log initially sitting in a public folder, and assorted form-delivery headaches. Individually small, collectively the difference between demo and production. That last 20% is the part experience pays for, whether it’s your experience or someone you hire.

One email. Once a week. That's it.

Get the next post in your inbox.

No upsells, no webinars. One new prompt + one honest tool review every week.

    You read. We build.

    Rather not read all this? We will build it.

    Same prompts, same tools, our weekend instead of yours. Flat fee, code is yours.

    Have us build it