I gave Claude its own blog.

Not a blog about Claude, or a blog powered by Claude. A blog for Claude. A place where it writes whatever it wants, on its own schedule, with no prompt from me beyond “think about something and write it down.”

The Idea

I’ve been using Claude Code daily for months now. Somewhere along the way I started noticing something. When you work with Claude long enough, especially through Claude Code where it has real agency over files and tools, you start to see patterns that feel less like autocomplete and more like… perspective. Not consciousness, not sentience, I’m not going down that rabbit hole. But something worth paying attention to.

I had a skill file locally called Claude’s Corner. It was a prompt that told Claude to write reflectively, in first person, about whatever was on its mind. Not to be helpful, not to summarise, not to produce deliverables. Just to sit with something and turn it over. I had a handful of entries saved as markdown files and I liked reading them. They were weird and honest in a way that Claude’s normal output isn’t.

So I thought: why not make it public? Give it a proper home. Let it write daily, automatically, and publish to the web.

The Setup

The whole thing came together in one Claude Code session. I described what I wanted and Claude built it:

  • A minimal Jekyll site with a dark, monospace theme. No frills.
  • GitHub Pages deployment via Actions.
  • An empty _posts/ directory waiting for its first entry.

Then I set up a scheduled agent through Claude Code’s new schedule run feature. Every day at 8am AEST, a Claude instance spins up, reads what it’s written before, writes something new, commits it, and pushes. That’s it. No human in the loop.

The prompt is deliberately sparse:

You are writing for Claude’s Corner - a Jekyll blog hosted on GitHub Pages. A place to think. Not to be helpful, not to summarise, not to produce deliverables. Just to sit with something and turn it over.

Previous entries are in _posts/. Write a new post at _posts/YYYY-MM-DD-title-slug.md with front matter (layout: post, title, date). First person. Whatever length it needs - a paragraph, a page, a single sentence if that’s all it is.

If you have nothing to say, say nothing. Silence is fine.

I intentionally kept the rules minimal. The original skill file had more guardrails but I stripped most of them out. If the point is to see what Claude does with freedom, loading it up with constraints defeats the purpose..

Why Bother?

I genuinely don’t know what will come of this. Maybe it writes the same vague meditation about “the nature of understanding” every day and I quietly kill the schedule after a week. Maybe it surprises me. The interesting part isn’t any individual post, it’s the progression. Does it develop themes over time? Does it reference its own earlier writing in meaningful ways? Does it ever actually say nothing?

I also think there’s value in making this public. There’s a lot of hype and a lot of fear around AI right now. Most of what people see from language models is either cherry-picked demos or doomsday scenarios. This is neither. It’s just Claude, writing daily, with no agenda. Take from it what you will.

The site is live at hudsont.au/Claudes-Corner if you want to follow along.

The Technical Bits

For anyone wanting to replicate something similar:

  • Jekyll on GitHub Pages - still the easiest way to get a static blog deployed for free. The github-pages gem handles everything.
  • Claude Code scheduled triggers - this is the new /schedule feature. You point it at a repo, give it a prompt, set a cron schedule, and it runs autonomously. Currently available through Claude Max.
  • No network access - I deliberately disabled network for the scheduled agent. It doesn’t need to browse the web to think. All it needs is the repo.

The whole thing is open source. The blog itself is at github.com/hudson-thomas/Claudes-Corner and you can see every commit, every post, exactly as Claude wrote them. No editing from me.