I keep coming back to one idea: automating my blog with Obsidian, Cloudflare, and task-specific AI agents. This post is the working version of how I think about it right now — concrete, unfinished, and honest about the trade-offs.

The problem — ideas die in notes apps; generic AI forgets your rules every session

On the problem — ideas die in notes apps; generic AI forgets your rules every session: the short version is that the obvious approach mostly works, but the interesting parts are in the edges. I tried the simple thing first, watched where it broke, and kept the parts that earned their place. No grand theory here — just what actually held up when I shipped it.

Architecture in one diagram

On architecture in one diagram: the short version is that the obvious approach mostly works, but the interesting parts are in the edges. I tried the simple thing first, watched where it broke, and kept the parts that earned their place. No grand theory here — just what actually held up when I shipped it.

Obsidian vault = second brain

On obsidian vault = second brain: the short version is that the obvious approach mostly works, but the interesting parts are in the edges. I tried the simple thing first, watched where it broke, and kept the parts that earned their place. No grand theory here — just what actually held up when I shipped it.

blog-skill/ = agent rulebook (voice, frontmatter, publish steps)

On blog-skill/ = agent rulebook (voice, frontmatter, publish steps): the short version is that the obvious approach mostly works, but the interesting parts are in the edges. I tried the simple thing first, watched where it broke, and kept the parts that earned their place. No grand theory here — just what actually held up when I shipped it.

blog/ideas/ = backlog with status tracking (idea → ready → published)

On blog/ideas/ = backlog with status tracking (idea → ready → published): the short version is that the obvious approach mostly works, but the interesting parts are in the edges. I tried the simple thing first, watched where it broke, and kept the parts that earned their place. No grand theory here — just what actually held up when I shipped it.

That’s where it stands. v1 works, the rough edges are visible, and I’d rather ship and iterate than polish in private. More as I learn — follow along.