How I built this page

This site has needed a facelift for years. Not because the technology was outdated, but because every previous version of this blog eventually died. Quietly.

I’ve started blogs before. Many of them. They all followed the same lifecycle: excitement → a few posts → silence. Over the years, those genuine attempts quietly turned into a blog graveyard. Apparently, enthusiasm alone isn’t a sustainable publishing system.

What went wrong before

Before touching the implementation, I wanted to clarify the goals — mostly to avoid the same problems as before. Specifically:

• Publishing felt like work
• There was too much maintenance involved
• The workflow was too dependent on tools and clunky integrations
• Focus drifted away from the content itself

From that came a few design goals:

Minimal friction

Writing and publishing should have minimal friction. If publishing is annoying, it simply won’t happen.

Worth coming back to

The project should be worth coming back to. Not necessarily for readers, though that would of course be a bonus. Rather for me — encouraging me to refine it and write more. It should pull me back in, not push me away.

What I did different this time

Agentic

Instead of wiring everything together myself, I let AI agents generate most of the glue code.

With that my role becomes more reviewer than implementer: I write specifications, the agent proposes changes, and I approve them before they land. For this I’m using Claude, specifically the Opus 4.6 model through Claude Code.

I’ll cover that in more detail in a later post.

Tech Stack

The obvious choice would be to go with something like Wordpress, or some other hosted blog platform. It solves a lot of problems — mostly problems I don’t really have.

What I really want is:

• Writing in Markdown, since that’s already how I write most things.
• A minimal workflow. Publishing should be automated so I can focus on writing instead of logging into UIs, formatting posts, managing schedules, or thinking about SEO.
• Something that won’t break in six months.

I briefly considered building a small CMS, but quickly realized it would come with far too much maintenance overhead, thus jeopardizing the goal of making it worth coming back to.

Instead, I opted for a Static Site Generator, allowing me to write my content in Markdown, which I was already doing anyway. For me, the obvious choice was Hugo, as I already have experience with it and generally like working in Go.

Outcome

The first version took about an hour to put together — which is exactly the point. If publishing ever starts feeling heavy again, the whole experiment has already failed.