Hey,
This is for everyone who sees a post saying "I built a website with AI" and thinks you can now build an Umbraco site by pressing a button, the way people imagine WordPress works.
I just built my own site with AI. It took me four days. And I want to tell you how it really went, because the honest version is more interesting than the demo version.
It's not what it looks like
When you see "built in four days", it looks like a prompt turned into a website. It wasn't that. And it's definitely not the WordPress-style, pick a theme and click a few plugins, kind of thing that people imagine.
Also, be honest about what those four days were. A personal site is the easiest project there is. I was the designer, the developer, the editor and the client, all at once. No stakeholders, no content team, no legacy, and no deadline except my own.
Could you build a commercial website in a few days? It depends, of course. A very small, simple one, maybe. But a real commercial Umbraco build, with real editors, real content and real requirements? No. You know what I mean.
Who did what
- Claude helped me with the design.
- Cursor and plugins helped me build it in Umbraco.
- Me, in the chair, the whole time.
That last one is the part that doesn't fit in a screen recording. I was not watching something build itself. I was working, and AI was making me faster.
Days one and two: concept and design
I spent the first two days with no real code at all. Just the idea. Who is this site for? What should someone feel when they land on it? What should it say about me?
And it was back and forth. A lot of back and forth. Almost every round started with me saying "no, not that."
- The logo went through several versions. I wanted my name to be the logo, a bit like how some brands turn their name into a wordmark. First came a clean look, then a letter swapped for a symbol, then a slanted letter, then a custom-drawn one, and finally the pixel-block style that stayed.
- The headline had a typing animation that looked childish to me. I asked for something more serious, and we ended up with a fixed phrase and a line that fades.
- The Speaking section had to look completely different from the rest of the page. It became a set of ticket-style cards that open up with the details.
- And I cut things. I tried an illustration for my About section: first a cartoon of me, then windows and code, then something cleaner. I rejected every version, and today that section has none. A visual that doesn't earn its place is just decoration.
The AI never got tired of my "no". But it also never knew what "good" meant. It only knew what I told it. Taste, judgement and the final decision were mine, every single time.
And no, it doesn't look like it was made by AI
I can say this confidently: my site doesn't look like the usual AI-made site. You know the look. The same gradient, the same row of cards, the same layout that almost everyone gets when they type "make me a portfolio". Look at the site. The design alone doesn't scream "AI made this", and that wasn't an accident.
The AI gave me options, and I kept choosing differently:
- The logo isn't a font. The letters are drawn block by block, in a pixel style I picked after trying other directions.
- Visuals that didn't feel right were rejected and cut, instead of accepting the first thing that looked "fine".
- The design direction came from me before any tool touched it: what I wanted visitors to feel, and what I wanted to stand out.
- Anything that looked like a default, I said no to.
The tool is the same for everyone. The difference is how many times you say "not that".
Same tool, different taste.
Days three and four: building it in Umbraco
Then I opened Cursor, added the plugins, and started turning the design into a real Umbraco site.
Yes, it was fast. Things that would normally take me an afternoon took much less. It was a fantastic pair-programmer for the repetitive parts.
But I'm a technical person first, and that mattered more than any tool:
- I decided how the content should be structured, so I can edit my own site later without a developer.
- I read what was generated, and fixed or threw away what wasn't right.
- I knew which version of Umbraco I was on, and what that means for the code.
- I tested it myself, in the browser, on the real thing.
Whenever something didn't look right, I could tell why. That's the difference between using AI and being driven by it.
If this were a commercial site
My own site is the one place where I could afford to move fast. If I had to work on a commercial site, I would use AI very differently.
Where I'd use it: the base. Template slicing. Take the finished design and turn it into clean HTML, CSS, layouts and partial views. That is repetitive, well defined work, and AI is great at it. It saves real time.
Where I wouldn't: turning that HTML into the backoffice. This is the part you just can't automate. Think about what it involves:
- Deciding what editors can change and what they must not touch
- Choosing which things become document types, which become blocks, and which stay fixed
- Naming fields so they make sense to someone who isn't a developer
- Validation, defaults and what happens when a field is left empty
- Permissions, workflows and how it feels to use every single day
There is logic in it, and there is design in it. It's design for the editors. There's no single right answer, because the right answer depends on the people who will use it.
Slicing a template is a task. Designing a backoffice is a decision.
What the speed doesn't cover
Here is what is actually inside a website, even a personal one. None of it is exciting, and all of it matters.
- Content modelling. How pages and content types are structured decides how pleasant editing is for years.
- Security. Who can see and change what, and what is public.
- Performance, SEO and accessibility. None of it shows in a screenshot.
- Languages. My site is in English, Dutch and Danish, and every language version needs to be wired up properly.
- Analytics and privacy. Visitors in Europe expect their privacy to be respected.
- Hosting and deployment. Where it runs, how it updates, and what happens when it breaks.
- Editing experience. Someone has to make the backoffice friendly for the people who will use it.
- Accountability. If it goes wrong, a person is responsible, not a tool.
That was one personal site, with one editor, and no client. A commercial site has all of this and more people, more money and more risk.
Why this matters for Umbraco
Umbraco is a developer-first platform. There's no theme that decides things for you and no plugin that quietly fixes everything with a click. You get a powerful content model, full control of .NET and the freedom to build exactly what you want.
That's why people love it, and it's also why you can't prompt your way to a finished commercial site. Control means decisions. Decisions need someone who understands the consequences.
AI built fast. I made it right. Both parts were needed.
A message to three kinds of people
If you're a client: be careful when someone says a full Umbraco website was "built by AI in a day". Ask what happens on day two: security, hosting, handover, support. The first version is the easy part.
If you're a developer: use these tools. They are great, and they make a good developer much faster. Just stay in the chair.
If you're starting out: use AI to learn, not to skip learning. Ask it to explain things, then check the documentation. If you can't tell whether the answer is right, slow down.
So, can AI build a commercial Umbraco website by itself?
No, not by itself, and not in a few days. It can help a developer build one much faster, especially at the base, and that's already a big deal. But somebody has to design the backoffice, guide the tool, review the result, test it and answer for it.
Four days, a lot of "no, not that", and me in the chair. That's what it took, and I wouldn't have wanted it any other way.
Thanks for reading, Bishal