Visible Loops, Invisible Daemons

Everyone's building agents that run while they sleep. I want an army I watch breathe

I don’t want an array of crontab agents spawning and spitting out results. I want an army.

Everyone wants a second brain. The Jarvis sidekick, a superpower you control like a robot. The biggest cottage industry in AI right now is Claude Code plus Obsidian second brains.

Here’s why it works. The last ten years taught us that text files are the best format, because text is portable. Every knowledge-base tool that ran on plain text got adopted. That’s the rise of markdown. I picked Logseq, opensource, a little more niche than Obsidian, my markdown graph. Then you add an agent, and it just works. Agents are good at exactly one thing: grep and modify text. So the second brain gets intelligent the moment a driver takes the seat. Your agent.

People have been building this. But there’s a small step, in the wrong direction, that leads somewhere completely different.

That step is automation. And the fucking frenzy over it. The kids run Claude like cron jobs. A text message lands with the result, they tap, read, respond, and that kicks off the next cron job.

I can’t work like that. What I don’t like about cron and claude -p is that it’s invisible to me. I can’t see it.

I want the session alive, so I can connect to it when I need to, and debug it easy. I want my robots to persist.

I don’t start by building for things to run in my sleep. Watching it is the whole point. “Run this while I sleep” comes later, for the things I don’t care about, and honestly, that should be procedures. Pure code. Not protocols, where agents drive the behavior and the decisions.

So back to the army. Organized in a hierarchy, marching under my command, toward my mission, my goals, my fight. Each unit independent, merged, rested. Add generals and mini-kings when I need them. And all of it while I watch.

It’s way too fun to sleep while this works.