We moved borobit.com from WordPress on a rented server to AWS, as a static site.

I want to say one more thing at the start. This is not a text against WordPress. For people without a technical background I still think it is the better choice. The approach I describe is for teams like ours, who build software and treat their website as software. This way we can apply to our website the same processes we use when we build software.

Why AWS

The rented server was with one of the large Bulgarian hosting providers, and it ran borobit.com and a few more of our sites. The contract was ending, and we had to either sign a new one-year contract or move to another provider.

We didn't renew it, because we build cloud systems in AWS every day. I have written more about that in a separate article. It made sense for the website to be there too, so we decided to migrate all the sites from the rented server to AWS as static sites.

With the sharp progress of AI lately, I was already writing the new pages of the site with the help of AI anyway, and only pasting them into Elementor as HTML. So migrating was the natural next step. There are other benefits too, like the cost and the speed, and I talk about them further down.

How the new site is built

The new site is built with Astro. The pages are assembled in advance, and AWS holds ready-made HTML files. There is no database and no admin panel. When someone opens a page, it is sent straight from the nearest AWS location. The files are in Amazon S3, and they reach people through CloudFront, Amazon's content delivery network, which has servers in Sofia too. All the infrastructure around the site is described in code with Terraform. Releasing new things or changes to the site is automated, with one test environment and one production environment.

How we moved it

I wanted to describe the whole migration process and share it in this article. So I told the AI to keep a journal of every step, mistakes included.

We wanted the site to move to AWS exactly as it looks, with no changes for visitors during the migration, so that we wouldn't mix things up. That also meant moving the DNS properly.

Before we stopped the old site, I told the AI to describe everything it does, not just the pages. That surfaced a few things that are easy to miss. The DNS of three of our domains that only redirect was on the server itself and would have stopped with it. The server also held two files that phones use to recognise links to the BoroBit App, and without them those links would have stopped working. And every subdomain of borobit.com pointed to the server's address. Once we released it, the address could go to another customer of the hosting provider, and our subdomains would lead to someone else's server. So we removed those records.

The AI also took pictures of the whole old site, every page on a computer and on a phone. The pictures were there so the AI could make the new site the same and compare it against them. Of 160 pictures, 156 differ by less than 1%. The titles and descriptions for Google match the old ones on all 71 pages, and every old address either works or leads to the new one.

The whole move took two days. The AI carried over the pages themselves in about two hours. After that I had it rework the whole structure two or three more times, so the site can keep evolving for many years.

To switch to the new site, once we had built it in AWS, we simply changed the DNS records.

Speed

We measured the old and the new site the same way, with Google PageSpeed Insights. Five pages, each five times, taking the middle result.

On a phone the old site scored between 96 and 100 out of 100, and the new one scores between 99 and 100. On a computer both score 100. On a phone the main content appears sooner on all five pages. On the home page in 1.5 seconds instead of 2.1, and on the article about bank fees in 1.6 instead of 2.4.

If you only look at the scores, the difference is small. The old site scored well because WP Rocket didn't load a large part of the JavaScript until a person moved the mouse or touched the screen. Google's test doesn't move the mouse, so that part doesn't count towards the score. But it was still there, and the browser downloaded it on the first touch. The new site doesn't have it.

Home page Before Now
HTML 511 KB 95 KB
Scripts 54 4
JavaScript waiting for the first touch about 1 MB 0

Sizes are after decompression.

The first version of the new site was actually slower than WordPress. In the new structure we had carried over a lot of the WordPress legacy, heavy styles, fonts and animations waiting for JavaScript. That is why, as I wrote above, we reworked it two or three times until we got it right.

Cost

The server cost about 100 euros a month excluding VAT, and it also ran other sites, which are moving to AWS as well. There the site costs us pennies. A site with our traffic fits within CloudFront's free monthly allowance, which is 1 TB of traffic and 10 million requests. We pay for the DNS zones, half a dollar a month per domain, and a little for the files and for assembling the site on every change.

The short version for the readers of BoroBit App, with the numbers for each page, is on the page How this site is made.

What the AI did and what I did

A large part of the work was done by AI. One AI wrote the plan and the infrastructure, and the other, Claude Code, wrote the site's code and the tests. I set the direction, made the decisions, reviewed, and applied every change in AWS myself.

The review took longer than the mechanical move itself. The first time I looked at the site on my phone, I found four things. The page jumped when you reached the bottom and went back. The menu button was off to one side and turned black when pressed. You couldn't change the language from the open menu. And the menu itself opened rather stiffly. Things like that don't show on a picture of the page, only when you hold the phone. Once we found them, I told the AI to write tests for them too (actually, I didn't tell it explicitly, it's in the main instructions for the site).

Then I asked for the site to feel smoother. The look stayed the same, but moving between pages is now smoother and the menu is no longer stiff. There is still room for improvement here, but that is a matter of separate work, outside the migration.

The AI was fast, but it made mistakes, and you have to be very careful with it. It assumed one of the domains had no email and would have switched it off. Twice it broke the styles while cleaning up unused CSS. A CloudFront function passed the local check but didn't run in the real environment. And in the first version it put the redirects of the old addresses into the infrastructure (instead of the site itself), so every change to the site would have required a change there too.

None of these mistakes reached the live borobit.com. They were caught by the tests, by the check before every change, or by a question of mine asking it to explain a specific action it took on a task. The infrastructure itself is set up in code, there is a test version, everything runs automatically and only then goes to production, and if something breaks, we roll back. The site is now developed as a software project, and that is really why we migrated to a static site in AWS.

What we lose

We lose the visual editor. Without code or AI, a person without a technical background will find it hard to change anything on the site. For us this is not a problem, because we maintain the site ourselves, but for many companies it is.

We also lose the ready-made plugins. Without them there is nothing to update and no third-party code with vulnerabilities on the server, but we also lose everything they give out of the box. If we need a form, a calculator or a shop, we will have to build it ourselves (though with AI that is now the easy part of the puzzle).

You also need someone who understands what the AI does and can check it.

That is why this approach is for technical teams. It doesn't fit when the site will be maintained by someone without a technical background, or when it relies on plugins, for example for a shop or bookings.

What we gain

We can now develop the site for years without fighting the platform. We publish new content quickly and the right way, and it reaches people as fast as possible. We can look after the SEO, the speed and every part of the site separately. The struggle with two languages is gone too. Before, I translated in the plugin section by section, paragraph by paragraph, and automatic translation cost extra. Now every page has its Bulgarian and English version side by side, and the links between them are made automatically.

We are a software team and we maintain the site the same way as our product. The same goes for our other sites.

Until recently, if your site had no visual editor, every change needed a developer, and often a designer to decide how it should look. In a small team, these people's time goes to the product and to clients. That is why WordPress and Elementor made so much sense. With the progress of AI this has changed. The code and the design changes are done by AI, and I say in words what I want, review it and release it. We move one level up. We no longer make the change ourselves, we decide what it should be and check it.

Petyo Stoyanov
Inabit
petyo.stoyanov@inabit.eu