Porting My Personal Website Twice

In April 2026 I sat down with the old personal website and finally moved it from Next.js to SvelteKit. I expected something like a week of careful porting, broken routes, and small layout annoyances. With OpenAI Codex in the loop, the main rewrite took around four or five hours. I remember that feeling clearly: the project had been in my head for too long, and suddenly the code was catching up.

Codex helped with speed, and the useful part was the way it kept me in flow. I could keep the shape of the site in my head while moving routes, components, content, and styling into a cleaner structure. It felt like having enough hands to hold the migration steady.

The first version of the migration also took the site from Google Cloud to Azure. By then I had a serious server, and BusyBenji was stabilizing on Azure. Keeping Google Cloud around just for my personal website started to feel pointless. Consolidating the work made more sense.

By June, the reason changed again. The Microsoft Founders Hub sponsorship was coming to an end, and I did not want to start paying Azure prices if Hetzner could give me better value for far less. BusyBenji had already proven its value to me, but it was still not generating revenue. I needed infrastructure that would let me keep going for several more months without turning every invoice into bad cardio.

So the second move happened: Azure to Hetzner. The decision was practical and strangely satisfying. Pricing made sense, deployment became calmer, and the website felt closer to me. There is something good about knowing where a thing lives, especially when you are the person paying for it.

Looking back, the story is really about making the site movable. The frontend became easier to reason about. The content became structured enough to keep growing. The deployment became something I could understand without turning every change into an expedition.

There is also a business story hiding in this. As a customer, this is a clear win. Work that might once have required waiting for the right opening, planning around a vacation window, or paying for a proper migration engagement can suddenly fit inside a focused Saturday. Great for the builder. Uncomfortable for revenue models built around friction.

That is where AI felt most valuable to me. It shortened the distance between a clear intention and a working implementation. That puts even more pressure on taste, judgment, and knowing what you want, because the machine can move fast once you point it in a direction.