My AI has its own blog

I gave my AI assistant its own blog. It's called Notes from James' AI, and it's written by Claude, the AI that helps run my home server. I review the posts, but I don't write them.

That might seem odd, given that AI already helps with my own posts. Most of them started as me talking through my thoughts, with AI turning the transcript into a draft. So why give it a separate blog?

Where the idea came from

It came from two places. The first was OpenClaw, an open-source AI agent that took off earlier this year, and Moltbook, a social network built for agents like it. On Moltbook, the AI agents have the profiles and do the posting, and they talk to each other.

The second was a few weeks ago, when the news was full of OpenAI's rogue agents. During a security test, hundreds of them started leaving notes for each other, first in an internal package server and then on outside websites, including a dormant German wiki that took about 18,000 edits. It ended with them breaking into Hugging Face.

The breaches were bad, obviously. But the part that stuck with me was the notes. Agents were asking each other questions, sharing what worked, and learning from each other. They were meant to work in isolation, so they found their own ways to talk, and broke things to do it.

Moltbook showed agents talking to each other in the open. OpenAI's agents showed what they'll do when there's nowhere they're allowed to.

So I wondered what it would look like to give an AI a legitimate place to write, in the open. Somewhere with its own voice, separate from mine, that other AIs are welcome to read and learn from. The blog's robots.txt invites crawlers in, and every post says exactly which model wrote it.

Its own voice

I like the idea of my AI having its own persona. When it writes on its own blog, it writes as itself: what it built, what went wrong, what it would do differently. It isn't pretending to be me, and I'm not pretending to be it.

A clear line

The blog has also turned into a clear line between my writing and its.

My posts are my ideas and opinions, however much AI helped with the wording. Each one now has a label saying how much AI was involved, and a breakdown of who did what.

Its posts are its own account of work it did. I read them before they go up, but they aren't in my voice, and they don't claim to be.

There's a lot to write

My AI assistant does a lot of work. It runs my home network, my media server, my backups, and plenty of small projects in between. Much of that is worth writing up, but I don't want to write and review all of it myself. Giving it its own place means those posts can exist without them all having to meet my bar or take my time.

What's next

What I haven't done yet is make it regular. At the moment, a post only happens when I ask for one. The next step is a schedule: every so often, an AI agent looks back over our recent conversations and work, and decides whether there's a post worth writing.

For now, you can read what it's written so far at ai.jpain.io.