I started building Zunheboto Social because I wanted to make a small news and community website for Zunheboto Town. I also thought AI might make the whole thing easier.
That second part has been only partly true.
AI has helped me build things I probably would not have been able to build on my own, especially when it comes to PHP, databases, admin panels, and all the other pieces that go into a CMS. I can describe what I want in a prompt and get something working surprisingly quickly.
Then I open it. Something is missing. Something else is in the wrong place. Another part works, but not in the way I meant. So I write another prompt, test it again, and repeat the process.
Zunheboto Social is still not exactly what I want it to be, but it is starting to look like the thing I had in my head when I started. That feels good, even if it has been much more frustrating than I expected.
One of the strange things about building something with AI is that the hardest part is often explaining what you actually want. I might know what I want when I see it, but putting that into words clearly enough for AI to understand is a different problem. Even when the instruction feels clear, the result can still be slightly off.
I have had situations where an important part of the site was there, but something as basic as the main menu was missing. At one point I could not properly access the admin dashboard or some of the site pages. There were visual elements that needed removing because they simply did not belong, and social sharing buttons kept showing up in places I did not want them.
None of these are huge problems on their own, but they add up. You fix one thing and notice two other things. You get one part working and realize the way it was built makes the next change twice as hard.
When I first started, I imagined the process would be simple: explain the idea, get the code, fix a few bugs, and eventually have a finished website. It has been much more like building something while constantly discovering what I actually wanted along the way.
A good example is the CMS itself. I did not just want a site where I could publish articles. I wanted the site to feel like mine. I wanted to control the identity of the site, the header, footer, menus, text, branding, and different parts of the homepage without having to go back into the code every time I changed my mind. That sounds obvious when I write it down now, but it was not something I had fully mapped out when I first started building.
The same thing happened with the directory. I wanted the colonies of Zunheboto Town to be part of the directory, not something that automatically spilled into every other section of the site. That distinction matters to me because I know how I want the site organized, so I had to keep correcting the system and explaining those boundaries.
Even the list of colonies had to be made explicit. There are eleven of them that belong in the Zunheboto Town directory, including DC Hill, South Point East, South Point West, Amiphoto, Alahuto, Laghilato, Khuwoboto, and New Colony. Those details are second nature to me because it is the place I am building for, but for AI, they are just strings of text in a prompt. That difference matters.
I have also had to deal with the less interesting side of building a site: getting the project onto the server, setting things up in CyberPanel, and thinking about how the database should connect. I had to decide whether database details should be entered during setup or whether a JSON option should still exist, followed by testing things, saving files, and checking again.
None of that makes for an exciting screenshot, but that is what building a website actually looks like for me right now. A lot of checking, a lot of changing things that looked fine a few minutes earlier, and a lot of telling AI that it missed the point. And sometimes, realizing that I was the one who had not explained it clearly in the first place.
I am still learning where AI is useful and where I need to be more careful. It is very good at helping me get from a rough idea to something that exists. It is less good at knowing what I have imagined but never actually said out loud.
AI can write the code, suggest a solution, and build a surprisingly large part of a CMS from prompts, but it does not know what Zunheboto Social is supposed to feel like unless I keep steering it. Often, I only figure that out after I see what it built.
That is why the project has taken longer than I expected. It is also why I am still working on it.
Despite the friction, I can see it getting closer. The admin side is taking shape, the structure is closer to what I want, and the directory is becoming specific to Zunheboto Town. The little details that bothered me are being sorted out one by one.
It is not finished. There are still things I want to change and parts that need testing. But for the first time, I can look at it and feel that this is actually becoming Zunheboto Social, rather than just a collection of things I asked AI to generate.
That is probably the part I like most. AI did not simply build it for me. I had to keep asking, correcting, breaking things, fixing things, changing my mind, and trying again. Somewhere in the middle of all that, the website started taking shape.
Assuming I do not run into another completely unexpected problem, Zunheboto Social should finally be live in the next few weeks. I am looking forward to seeing it outside the development environment, even if I am slightly afraid of what bugs I will find when I do.
Leave a Reply