Around 2010, if a small business needed a website, the answer was obvious. You went to ThemeForest, paid 40 dollars for a theme with a beautiful demo, installed WordPress, and felt like a genius. Launch day was magic.
Then you lived with it.
Your content was wired into the theme's shortcodes and page builder. Switching themes turned your pages into a soup of broken [vc_row] tags. The theme bundled a dozen plugins, updates broke things, and the developer who customized it moved on. Three years later you paid someone to rebuild the site, and the cycle started again.
The industry learned the lesson slowly and expensively: the problem was never WordPress. The problem was that content and presentation were fused together, and the presentation layer was a pile of dependencies someone had to maintain forever.
We are doing it again, faster, with nicer tooling.
The new 40 dollar theme
Today you type a few sentences into Lovable or v0 or Bolt and a real website appears. It looks better than any ThemeForest demo, and it took ten minutes. But look at what was produced: a React application, with a build pipeline and a thousand transitive dependencies. And your content, the entire point of the site, is string literals inside JSX files.
Run the tape forward. The owner wants to change their business hours. That sentence lives in a component, so they prompt the AI and hope the diff only touches what they asked. They want their marketing person to handle the blog. The marketing person looks at the repo once and never again. Eighteen months later the framework has a new major version, half the packages are deprecated, and the build needs an afternoon of dependency surgery from someone who charges by the hour.
The shortcode soup is back. It just compiles now.
Why this keeps happening
Tools optimize for the moment of purchase, and the moment of purchase is launch day. A ThemeForest demo was engineered to win the 30 seconds you spent browsing previews. An AI builder is engineered to win the first ten minutes of prompting. Neither moment contains any information about month ten, so month ten is nobody's problem. Maintenance does not demo well.
There is also a confusion both eras exploit: the difference between a web app and a website. A web app is software, with logins, state, and user data. React exists because apps genuinely need it, and AI coding tools are legitimately great for building them. But a website is mostly content with a layout, and the durable technology for that has been HTML for thirty years. HTML has a property almost nothing in computing has: it does not rot. A page written in 1996 still renders. No framework can promise that, because frameworks are alive, and living things need feeding.
Use an app-shaped tool to build a content-shaped thing, and you inherit app-shaped maintenance for something that never needed it.
The test I would apply to any site builder
Not "how good is the first draft." Every tool passes that now. Instead:
- Can a non-technical person fix a typo without touching code or trusting a diff?
- If you add a tenth case study, is it guaranteed to match the other nine?
- Can someone get blog access without the ability to break the site?
- Will the site still build, unmodified, in five years?
The ThemeForest generation failed all four and we paid for it for a decade. Most AI-generated sites fail all four today. The demos are so good that nobody is asking.
You are reading this on the SleekCMS blog, so I am obviously talking my book: we built an AI site builder where the content lands in a CMS and the templates are plain HTML, precisely because of the years I spent inheriting other people's launch-day magic at my agency. Judge the argument on its merits. But if you got a site out of an AI builder recently, open the repo and find the sentence with your business hours in it. That is where you will be living for the next few years.