This blog
I haven’t paid much attention to this blog for a while. Now as I try to refocus, I decided to change that and write a bit more.
I also thought it might be a good opportunity to try some of the new AI coding agents and see it they can help me redesign and modernize this site. I had great experience coding with Claude in the past, but OpenAI Codex was just released so I thought I’d give that a try.
As for the new design, I recently came across Phil Schmidt’s blog and loved the simplicity. Full credit to Phil’s design, this site is heavily inspired by his.
Hosting
- Astro seems to be the most popular JS static site generator right now, so I went with that. I was able to achieve almost everyting I wanted, but it was not as smooth as I hoped. Honestly, I found Hugo a bit cleaner and simpler to use.
- I’m hosting the blog on Cloudflare Workers. I thought I started with a Cloudflare Pages Astro template (via C3), but turns out those framework templates now use Workers under the hood, which is nice. Once the CI/CD integration from Pages makes it into Workers, I guess there’ll be little reason to ever use Pages again.
- I am fully sold on the serverless idea. Not having to worry about infrastructure is freeing and allows one to focus on the things that actually matter.
OpenAI Codex
- The CLI is a bit limiting, things like multi-line entry are not easy and referencing images with say screenshots can be done, but it awkward.
- What didn’t work for me was asking for complex large tasks.
- Instead, a more successful strategy was small incremental tasks (“now add a dark mode”, “add estimate for reading time”).
- CSS was a hit and miss, I tried sharing screenshots of the desired end state, I tried describing it in DOM-terms, and oftentimes I just had to do it myself.
- On the other hand Codex was surprisingly adept at working with Astro and even discouraged me once from trying to achieve something, explaining that this goes against Astro’s best practices.
- I initially tried approving every change, but quickly switched to
--full-auto
mode and just monitored changes from the last commit in git.
My total spent on Codex for this blog was $8.27. And despite the experience not being anywhere near flawless—often requiring multiple attempts for simple tasks and occasionally hitting dead ends that needed manual fixes—it still saved me a significant amount of time. So I’d say overall it was worth it and helped me build on Astro without having to spend too much time researching its oddities.