Agent Makers: From a Personal Site to a Structured Publishing Experiment

When I started spending more time building things on my own with coding agents, one of my first thoughts was pretty simple: I should probably have somewhere to write about all this.

So I started Agent Makers.

At first, I mostly wanted a site of my own — somewhere I could keep writing about things I was learning, experimenting with and building. I also liked the idea that, over time, a few friends and like-minded people might contribute too and it could become a small community rather than only my personal site.

And, of course, I wanted somewhere to publish the projects.

The slightly funny part is that once I started building the projects, I spent a lot more time building than writing about building. So the articles slowed down. The humor pieces slowed down. The site just sat there waiting for me.

Agent Makers is still technically quite simple: Next.js, React, TypeScript, Tailwind and filesystem-based content, deployed on Vercel. There is no CMS or database behind the publishing side.

But I kept adding little conveniences. Paste a video link and it can become an embed. Put several images together and the site can turn them into a gallery. Multiple videos can become a grid. A plain link can turn into a preview card.

So one idea gradually emerged: write normally, and let the system figure out some of the presentation.

When I came back to the site to publish the recent Project and Humor pieces, I ran into a different problem.

Those kinds of articles have some structure of their own. A project story is not just arbitrary prose. And a humor piece definitely should not be forced into the same shape as a project story.

So instead of authoring those directly as Markdown, I started describing them in structured YAML and generating the final site content from that. Projects now have one grammar. Humor has another. Both go through validation and generation before becoming the Markdown the site renders.

That gives the site two slightly opposite ideas: prose → infer presentation and structure → generate the publishable article.

I did not plan that when I first built it. It just emerged from using the site.

How I build changed the same way: code in a chat window, pasted in an area at a time. Then generating inside the editor, a program at a time. Now mostly a specification, and an agent working from it.

The Project and Humor pipelines are really only a start.

Even a normal blog article can mean very different things. It might be a viewpoint, a personal narrative, an explainer, an experiment log, or a research synthesis. Those are not just different topics. They have different semantic shapes.

So one thing I want to explore later is whether each kind of writing can have its own lightweight structure and extraction or authoring process — while still leaving room for the writing itself to feel natural.

For now, Agent Makers is still mostly my little publishing site. It has not become the community space I originally imagined, and I certainly have not written as much as I thought I would.

But it has become something else I find interesting: a place to experiment not only with what I publish, but with how publishing itself can become a little more structured and programmable.