cloud / systems
It started with Kraftwerk, my own strength-training app. I had designed it with Codex. Afterwards I noticed that my old WordPress site in my AWS account was no longer running. I had owned the domain berndkuhlen.de for a long time. So: why not try to rebuild the website with Codex as well?
My first question was a very practical one: which architecture is cheap enough for this? Codex suggested S3 with CloudFront. For my personal site I took that route instead of running containers on ECS with a load balancer.
Switching the old site back on
I started the old EC2 instance with WordPress under berndkuhlen.de one more time – only to read the content out. In parallel, the new site was built temporarily under new.berndkuhlen.de.
Only after I had checked that my articles had been carried over did I shut down the old EC2 instance, the WordPress subscription and the load balancer. Then the new S3/CloudFront site moved to berndkuhlen.de, together with a new certificate from AWS Certificate Manager. My training courses and certifications were carried over from my LinkedIn account.
I have already described the technical migration in a longer article. Here I am more interested in how all of it developed in everyday use.
After that, one wish followed the next: show visitor numbers. Add Kraftwerk and share my training progress – bye bye, running app … ;) Integrate the games I had already built with Codex. Write new articles and translate the content into more languages.
All of it through prompts. I have not written a single line of code myself. What has been laborious so far were mainly frontend bugs: something looked different on one device or did not work in one browser. Then it was a matter of opening the dev tools, hunting for the error and handing the findings back to Codex.
The quota slows things down as well: with my OpenAI subscription at around 23 dollars, the tokens were often used up faster than my ideas. Then I had to wait – or spend additional money.
A lesson for 15 US dollars
In the beginning I had no sensible cost control in place. A global invalidation of the CloudFront URLs then cost me around 15 US dollars. For this small website, a fairly annoying surprise.
By now a cost check comes before every change. Cache invalidations are fundamentally forbidden in my projects. The objects are allowed to use their normal cache lifetime; not every change has to be visible everywhere immediately.
Meanwhile the site is also my little playground: easter eggs, more games and an AI-supported weather page. I can simply try out what is possible.