Google shipped an open format (OKF) for AI agents. My site already spoke it.

Google spent years selling Dataplex as enterprise data plumbing nobody outside a data team ever thought about. Last month it quietly became Knowledge Catalog, which Google now describes as an “always-on context engine” for AI agents.

The product underneath barely changed. The pitch changed completely, and that’s the tell. When Google relabels its data catalogue as a feeder for AI agents, it’s telling you where it thinks the value sits now.

The part I actually care about shipped in the same launch. They published OKF, the Open Knowledge Format. Strip the branding and it’s almost aggressively plain. A folder of markdown files, each with a few lines of YAML at the top, linked to one another, that an AI agent can read straight off without scraping your HTML or calling an API. Nothing to install and nothing proprietary holding your content. You could write one in a text editor.

Here’s why it’s worth a note. That’s the shape my site has been in for a while. My posts are markdown with frontmatter. My notes are markdown with frontmatter. Google just put an official name and a spec on the thing I’d already bet the site on. So I converted the lot. Every published post and note, 51 of them, now sits at /okf/ as a proper OKF bundle, generated from the same files the site builds from and rebuilt on every deploy so it can’t drift.

But, let’s be honest, because it’s the same one as last time. Nothing reads OKF yet. No major model goes looking for these bundles, the spec is a v0.1 draft a few weeks old, and the repo itself admits it isn’t an official Google product. Measured today, my bundle changes nothing. This is the same move I made with EntityMap in May. Ship the infrastructure early, expect it to sit quiet, let it compound if the format takes. Schema markup needed most of a decade to earn out. I’ve made my peace with slow.

What’s different this time is who’s doing the formalising. EntityMap is a sharp little spec from people I rate. OKF carries Google’s weight, bolted onto a product they’re actively rebuilding around AI agents. When the company that runs the index starts calling your content “context” to be served to agents, the sane response is to serve it in the shape they’re describing, before everyone else clocks that they need to.

The converter that built my bundle is a script right now. I’m tidying it into a free tool so you can point it at your own site and get a valid bundle out without writing code. More on that shortly.

For now it’s live, it’s quiet, and it’s holding its spot next to my EntityMap file and my llms.txt. Another small bet on the same future.