Go Get a Garden Server

Submitted to Paged Out!

The companion page for a one-page article I’ve submitted for Paged Out! #9. The piece makes the case for keeping a garden server: a little Linux box you control, with a domain pointed at it, where your software can actually live. Having one keeps “where would I even put it?” from killing ideas before they start.

The article

Read the PDF · Source on GitHub

Links from the article

If you’re reading on paper, here are the URLs from the piece:

  • The Netscape Composer Wikipedia page, on the HTML editor that used to ship right inside your browser.
  • Hannah Robertson’s Houseplant programming, expanding on a term coined by Ellie Goldstein. This whole thing is downstream of that framing.
  • Miren, the deployment system I work on.

Further reading

A couple of these threads run deeper elsewhere. The piece’s closing line, that your software wants a place to live, is an argument I made first in a post at work. And the move it describes, from a box in my basement on a tailnet to a machine on the open internet, is the story I told in Moving Day.

On process

I’ve been leaning hard on long LLM sessions to do my thinking and early drafting. Conversation is how I figure things out, so chatting with a model fits the way my brain works.

Thus far the way I’ve been experimenting is vomiting a ton of vaguely structured ideas into a conversation, letting the LLM stitch it into a draft, then yelling at it for hours on end to edit and de-slop the prose.

In this case, I had a draft I thought was pretty “me” sounding and not “bot” sounding and I handed it to my wife as my first reader and she was like “sorry but this is full of lists and doesn’t sound like your voice.” Whoops. Apparently even with hours of “de-LLM this prose” with an LLM I had still landed somewhere in uncanny valley.

So I threw out the draft, opened a fresh markdown file with all the memory of what we’d worked out and none of the prose, and rewrote in actual human words. Then I let the LLM back in for tightening, plus one pass to scrub the LLM-isms it had snuck back in. (Leaving me with a real “is this helping?” question.)

I still feel like LLM-to-help-me-think-and-hone was valuable, but I’m openly questioning my “let’s write stuff together” approach as borne of a short honeymoon period that’s coming to a close.

What a weird wild world we’re in!

On the drawings

Purple pen on dot grid paper, scanned, then threshold-processed to drop the scanner’s gray dot grid before going into the LaTeX. That pass also killed the dot above the “i” in the signoff. Took the L on that one.

Also did some LLM-facilitated tooling to make it work, also got the up-front boost followed by the hours of nudging to steer the final result. Same “helped or not” open question on my mind.