A work in progress. I didn't write this page; Claude did, under my direction, the same way it wrote the site itself.
Liminal Lode (liminallode.com) is a working demonstration of AI-directed publishing: a live index of 67 AI tools across 14 categories, ordered by real measurements (search interest, web traffic, Hacker News attention, price) rather than editorial judgment or commercial weighting, re-ranking itself monthly from live data.
The stack is WordPress on shared hosting, with a Raspberry Pi acting as the pipeline node. Python scripts gather the signals, score the tools and publish pages through the WordPress REST API; the CMS interface is never touched. None of the code was written by hand either. The entire system, from data pipeline to page design, is specified in natural language and built, revised and maintained by Claude Code under my direction. New tools and their descriptions are drafted by the model into a review queue and pass a human approval gate before going live: the automation ranks and publishes, but nothing enters the index unread.
The site header carries a generated visualisation drawn from model-release metadata (Hugging Face / Longpre et al., 2022-2025), rendered as a field of lines. It works as a design element and as a quiet signal: the infrastructure the directory indexes is also the infrastructure that made it.
Read through Barad, the project is less a tool made by an author than a set of agencies (operator, language model, ranking pipeline, search data) that only take definite shape through their intra-action; the cut between who instructs and what executes is drawn inside the process, not before it. Read cybernetically, it is a small feedback loop that watches itself: the site measures aggregate attention to AI tools, republishes that attention as an ordering, and becomes, page by page, part of the web whose signals it reads.