Google just shipped a bunch of Markdown files and called it a format. And honestly, that’s kind of the whole point.

Last week Google Cloud dropped the Open Knowledge Format (OKF). I went in expecting some heavyweight knowledge graph spec with SDKs and runtimes. Instead it’s… just Markdown files in a folder. Each one has a bit of YAML frontmatter at the top for type, title, description, tags, and a timestamp, then normal Markdown below it. Files link to each other with regular Markdown links, which quietly turns the whole thing into a graph. They even recommend throwing in an index.md at each level so an agent (or human) can get the lay of the land without opening every file.

If you’ve used Obsidian, this feels instantly familiar. It’s basically your vault. Frontmatter, backlinks, folders that become a graph. Google themselves point to Obsidian, Notion, Hugo, and the whole LLM-wiki pattern as the inspiration. The lineage goes back to Andrej Karpathy’s idea of an “LLM wiki” — Markdown that a model can actually read, maintain, and update without getting bored.

The real headline isn’t that Google invented anything revolutionary. It’s that the thing a lot of us have been doing anyway turns out to be the right primitive for this next wave. Plain text outlives every fancy app built on top of it. No lock-in, no painful exports, no migration hell. That’s why some of us quietly run big chunks of our actual life out of a folder full of .md files.

🔗 Google’s announcement: https://cloud.google.com/blog/products/data-analytics/how-the-open-knowledge-format-can-improve-data-sharing

This lines up nicely with a book I finally pulled the trigger on: Tiago Forte’s Building a Second Brain.

I haven’t finished it yet, so this is more “what I’m expecting” than a full review. The core is his CODE method – Capture what actually resonates (instead of hoarding everything), Organize by actionability using PARA buckets (Projects, Areas, Resources, Archives), Distill notes down to their essence over time, and then actually Express it by making something with the knowledge. You only really know what you make.

The through-line is that David Allen idea: your mind is for having ideas, not holding them. Get the storage off your brain so the creative part can actually work.

Reading the OKF announcement and digging into BASB back-to-back felt like two sides of the same coin. Forte’s second brain is built for a human to think better. Google’s knowledge bundles are built for agents. Same Markdown underneath.

I’ve been wiring agents into my own Obsidian vault, and it works surprisingly well for text. Ask a question, it walks the notes, connects threads, and gives back something useful.

But as the vault grows, cracks appear. PDFs are clunky. Images are basically black boxes unless you pre-process them. That human “I know it’s somewhere near my notes” navigation by feel doesn’t translate to an agent. You start needing more explicit structure – a master index or menu note that tells the agent what’s there and where to look.

And yeah, that’s basically Google’s index.md convention. Different starting point, same conclusion: when you’re designing for agents, you stop relying on vibes and start drawing a proper map.

I’m still early on the “works at scale” part. Search that stays fast and accurate as the pile grows is still messy in my setup. But the direction is obvious.

These tools were built for one human riffing through their own notes. The next version needs to work for agents traversing them too. That means different assumptions about indexing, how to represent non-text stuff, and leaving clear trails instead of depending on muscle memory.

Obsidian nailed the basic primitive. Forte nailed the human discipline. Google just wrote a spec sheet for agents. The interesting work is in that messy middle layer — turning a pile of Markdown into something an agent can actually reason over reliably.

Peace out!