For the first time since the pandemic, I have resumed active, public development of Plerd, the software that freshly publishes this blog every time I update a certain directory of Markdown source files that lives on my laptop. I also use Plerd to publish Venthuffer, my videogame-monologue podcast, and ’Twas This, my “commonplace book” of shorter posts. Oh, and the Plerd News blog, which I’d literally forgotten existed prior to this week. But that’s where you can go read the details of Plerd version 1.9.

Now it can be told: while I worked full-time at Google between 2021 and 2024, the company’s legal department forbade me from working on Plerd, presumably because it had too much in common with Google’s own Blogger. (I had to clear all my personal, public projects with Legal, in fact, and Plerd was the only one they shot down.) Nearly two years after leaving the company, I find myself in a mood to blog more often, and I also feel more comfortable and confident with using LLM-based coding assistants, by way of my revival of BumpySkies.

Before I wrote any new posts, then, I identified some points of friction that I knew to discourage me from writing in public more often. With the help of Claude Code, I finished long-overdue improvements Plerd that addressed these issues, and subtly changed the way that I run Plerd’s software on my server. The latter of these call for some future documentation improvements; for now, I can happily share the results of the former.

First related thing: I also updated Web::Microformats2, a self-authored prerequisite to Plerd which quite possibly no project on earth besides Plerd uses. But it’s nonetheless lain quietly broken for several years, since one of its tests stopped passing, meaning that nobody’s been able to easily install Plerd onto any computer for a long time.

I fixed it, and also rolled in a bunch of changes suggested by Matt S. Trout, known to the internet as mst, who died far too young last year. It did pull me up short to see his open issue in the project’s GitHub tracker, unexpectedly letting me hear his inimitable voice one last time. Insofar as it’s appropriate to do so, I dedicated the release to Matt’s memory.

Second related thing: I have added “AI disclosure” sections to the README files of Plerd and Web::Microformats2, and to the Fogknife About page.

The About page update is the simpler one:

Every Fogknife article is wholly human-produced. I do not use generative AI tools such as ChatGPT or Claude to draft Fogknife article text. If an article does need to include machine-generated text, then that article clearly labels it as such.

I often illustrate Fogknife articles with art or photographs by others. To the best of my knowledge, none of these images make use of generative AI, unless otherwise marked.

I wrote last year about using generative AI while writing, sometimes. My stance towards this practice continues to evolve, and one way that it’s changed since last summer is believing far more strongly that to let an LLM-backed process actually dictate my writing in any way only flattens and weakens it. I may keep LLM-based tools close at hand for various purposes, but I do not let them speak for me.

The only exceptions I’ll ever make for this are certain cases of “legalese” and other text that’s closer to executable source code than ordinary prose—and even then, nothing gets published under my name without my full cognizance of it. And, indeed, that’s how I treat actual source code as well, which leads the sections I’ve added to my projects’ README files:

The following parts of this project include text that has been generated by LLM-based coding assistants:

• Source code, including comments
• Git commit messages
• Other developer-facing text, such as the CLAUDE.md file

All LLM-generated changes are human-directed and human-reviewed.

This README file, as well as all user-facing text and art assets, is entirely human-made. The software’s overall design remains solely my own, as does its ownership and authorship, aside from the human contributors mentioned above.