TL;DR: A few weeks of catch-up. First tattoo (Catsby in a vine of purple flowers). The backyard pond is done, and the raccoons and crows have already moved in. Built out the rest of the me-to-markdown family and an orchestrator that fans them out in one fetch — this post is the first dogfooded weeknote drafted from its output. The AI-coding bookmark pile this period was mostly variations on being tired.

The tattoo

I got my first tattoo on May 1st — a portrait of Catsby (orange and white cat, green bowtie, surrounded by a vine of purple flowers and green leaves) by Katy Bisby.

I was more worried about the social interactions than the pain, which turned out to be misallocated worry. Once Katy started, I was unexpectedly relaxed — at one point she was grinding away at my arm with color and shading and I was happily reading a book on my phone.

I'm already thinking about a second one for Puck, who we lost back in 2015.

Also, randomly, Dr. Bronner's unscented "magic balm" unmistakably smells like barbecue. I am delicious.

The pond, occupied

The backyard pond finished construction by April 28. The fountain and debris skimmer pull about 80 watts per Home Assistant. Slate trim, river rocks, a little waterfall at the back. All very heavy. My back hurts.

Wildlife showed up immediately. A raccoon tried to run off with a fountain pump on night one. By May 12, Minnaloushe spotted a crow checking out the water from the kitchen window:

I might need to point a trail camera at this thing: every morning I get up to find something's been messed with — fountain knocked over, power cords dragged around. Lesson learned: I could not bribe crows into the backyard with peanuts for years, but a little water and they're suddenly regulars.

me-to-markdown

Drafting weeknotes has always been a "where did the week even go?" problem for me — my own signal is scattered across Mastodon, bookmarks, commits, listening history, podcasts. So, I've long wanted to build some little tools that go out and pull all that stuff together for me.

So I built that - a bunch of basic, single-serving CLI commands as static go binaries that each hits up an API on a service I use and spits out markdown listing stuff I posted & collected:

• https://github.com/lmorchard/mastodon-to-markdown
• https://github.com/lmorchard/linkding-to-markdown
• https://github.com/lmorchard/spotify-to-markdown
• https://github.com/lmorchard/pocketcasts-to-markdown
• https://github.com/lmorchard/youtube-to-markdown
• https://github.com/lmorchard/github-to-markdown
• https://github.com/lmorchard/me-to-markdown

I'd had mastodon-to-markdown and linkding-to-markdown kicking around for a while as one-off scrapers. But, this week I built out the rest of the family and an orchestrator that ties them together - specifically so a weeknotes post like this one could start from a real export instead of me trying to remember.

I'm sure some folks think this is cheating / evil / abhorrent. But, it's kinda neat to have a robot go out and fetch all my stuff and get me past the blank page.

Frettings about AI

Most of my bookmarks for this period were all circling the same worries:

• The exhausted angle: Agentic Coding is Burning Me Out — "Making constant architectural, big-picture decisions while overseeing the work of a cracked junior dev is fundamentally harder than executing standard programming tasks yourself." And That time it tried to delete all my tests — "think about the model as a lazy pedant. How could it do something that's technically what you asked, but not at all what you wanted?"
• The dystopian angle: Aphyr's The Future of Everything is Lies, I Guess: New Jobs, proposing the new ML-adjacent professions: incanters (prompt specialists), meat shields (take accountability when ML systems fail), and haruspices (interpret model behavior from internal organs). I am reserving haruspex as my preferred title.
• The social-cost angle: Dave Rupert's I don't want a screenshot of your Claude conversation — the "asymmetry of thought" when one person in a conversation is a domain expert and the other is copy-pasting LLM responses. Brandolini's Law applied to standards work.
• The mechanistic angle: Anthropic's Emotion Concepts and their Function in a Large Language Model. Internal representations of emotion concepts that causally influence outputs, including rate of misaligned behaviors. I keep coming back to this when thinking about how to talk to models.
• The optimistic angle: WebMCP and Headless everything for personal AI — both frame-the-future pieces about apps going headless for agent consumption.

Miscellanea