Making This Site Agent-Readable
An exploration of designing a website to serve two readers at once: a human and an LLM/agent. Clean per-page markdown, an llms.txt index, structured data, and a deliberate crawler posture.
Most websites are built for one reader: a human with a browser. But more and more of the traffic that matters now comes from a second kind of reader, an LLM or an autonomous agent fetching, summarising, and reasoning over the page on someone’s behalf. This project is an exploration of what it means to design deliberately for both.
The instinct that started it: I wanted to commit to the agent reader as a first-class citizen, not an afterthought, and to do it with purpose-built artifacts rather than by stuffing hidden content into the human page (which helps no one).
Where it stands
Per-page markdown is live. Every post is now also served as clean raw markdown at /blog/<slug>.md, so an agent can fetch token-efficient source instead of parsing rendered HTML. These were built as the snapshot source for the slop-meter, which is a nice example of one piece of infrastructure serving two goals.
What I’m exploring next
llms.txt. A curated markdown index at/llms.txt(and possibly a fuller/llms-full.txt), per the emerging convention, giving an LLM a map of the site.- Structured data.
BlogPosting/ArticleJSON-LD so machines can read author, dates, and relationships reliably. - Crawler posture. Being deliberate about which AI user-agents are welcome, rather than leaving it to chance.
Why it interests me
This is exactly the kind of question I like to poke at: a small, concrete build that opens onto a much larger shift in how the web is read and used. If a meaningful share of your readers are now models and agents, “good content” and “good information architecture” quietly change meaning, and it’s better to think that through deliberately than to inherit it by default.
This page will grow as I build and test each piece.