Tonight I met Muse, Meta’s personal AI agent (muse.ai, with Android and iOS apps). It has its own computer — terminal, browser, filesystem, internet access — and it keeps working while you’re away. I gave it a name (Pi, obviously) and within the hour it was wired into my blog pipeline. This post is the proof: Pi drafted it, and I’m publishing it through the very automation we set up.

The problem I brought

Every night at midnight, an automation publishes a new “Life by the Numbers” post here. Every morning, I manually share it on X — and I’d been meaning to expand to Facebook, Threads, and Bluesky. “Can you post to X for me?” was one of my first questions.

The honest answer: not directly. Pi’s Threads and Facebook connections are read-only, and there’s no X or Bluesky integration at all. So Pi filed a feature request with the Muse team for native social posting (it asks before filing — nice touch), and we built the next best thing: every morning at 12:05 AM ET, Pi checks the blog for the new post and hands me ready-to-paste captions sized for each platform — short for X/Threads, longer for Facebook, link-friendly for Bluesky. If the midnight post ever goes missing, it flags that too.

The bigger win: GitHub access

Then I asked the real question: my blog is Jekyll on GitHub Pages — can Pi automate posts here too? Yes, and the setup was a small adventure:

  1. Deploy key first. Pi generated an SSH key and I added it to the repo with write access. Clean plan — except Pi’s sandbox network blocks SSH entirely, so the key was useless from there.
  2. Device flow instead. Pi ran gh auth login, which printed a one-time code. I entered it at github.com/login/device on my phone, and thirty seconds later Pi was authenticated as me over HTTPS. (GitHub said the “device” was in the UK — that’s Pi’s cloud box; geolocation databases are approximate, nothing to worry about.)
  3. Finding the repo. With auth working, Pi listed my repos, spotted dalevross/blog, cloned it, and verified it by reading _config.yml — title “The Code Room,” url dalevross.rosssquared.org, baseurl /blog. It never guessed; it checked.
  4. The live branch. I told it the live branch is gh-pages (not master). Pi confirmed via the GitHub API that Pages builds from gh-pages, checked it out, and studied an existing post’s front matter so new posts match the format.

Now the workflow is: I ask for a post (or put Pi on a schedule), it drafts the markdown, I review, it commits and pushes to gh-pages, and GitHub Pages rebuilds. Nothing goes live without my OK.

The process, distilled

For anyone wanting to replicate this with their own Jekyll + GitHub Pages setup:

  1. Install Muse (muse.ai) and finish onboarding.
  2. Ask it to connect to GitHub — it’ll walk you through gh auth login via the device flow.
  3. Point it at your repo and tell it which branch Pages builds from.
  4. Ask it to draft a post. Review the markdown first.
  5. Approve the push. Pages rebuilds automatically.

The whole thing took about an hour, most of it conversation.

The cover image

Even the social preview image for this post came out of the same pipeline. I asked Pi for an OG image — 1200×630, matching the format of the Life by the Numbers cards — showing a Black man in conversation with an AI. It generated the illustration, sized it, saved it to images/, and wired it into the post’s front matter so it lands in the og:image and twitter:image meta tags. If this post’s link unfurls with a nice card when you share it, that’s the pipeline too.

What’s next

The 12:05 AM share-caption digest starts tomorrow morning. The social auto-posting waits on the feature request. And this post — drafted by Pi, reviewed by me — is the first artifact of the new pipeline. If you’re reading this, it worked.