##  [Google has just named something I've been running for a year](/index.php/node/250) 

    *Submitted by Lennart on Fri, 19 Jun 2026 - 13:39*  

  ![Google has just given a name to something I've been running for a year](/sites/default/files/styles/wide/public/2026-06/composite_18.png.webp?itok=mjgU_yGD)

 

Last month, Google Cloud released an open specification called **Open Knowledge Format** — OKF. As I read it, I had a strange feeling of recognition. Because it describes, almost line by line, the way I myself have built up my knowledge base for over a year. Not because I was ahead of Google. But because the principle is old, and the direction the infrastructure is moving in now is the same direction a single consultant with a terminal also ends up in if they think it through.

This is the point I want to use this post for. Not because the format is exciting in itself — but because it says something about what AI readiness actually means for your company.

## What OKF is — briefly

Open Knowledge Format is Google's take on a common language for organizational knowledge. The core is surprisingly undramatic:

- **Ordinary markdown files** with a little structured metadata at the top (YAML frontmatter)
- **Organized in folders** with an entry file in each
- **Cross-links between the files**, so they together form a graph
- **Vendor-neutral** — it's a *format*, not a platform, a database, or a service

The problem they solve is one most companies know without having put words to it: knowledge is scattered. Some in the metadata catalogs, some in the wiki, some on the drive, some in the code, and a large part "in the heads of the experienced." Each time a new AI agent needs to be put to work, it has to gather that context from scratch. It's a waste, and it can't be moved.

Google's answer is to gather knowledge in one place, in one format, that both humans can curate and machines can read and update. They even quote AI researcher Andrej Karpathy: language models don't get tired, don't forget cross-references, and can touch fifteen files at once.

## Why it sounds familiar

This is precisely the architecture behind my own knowledge base. I write everything in markdown, with metadata at the top, organized as a graph of links rather than a folder hierarchy. A small tool called IWE makes the graph navigable, and a language model can fetch the exact right part of it when it needs to help me with a customer meeting or a blog post. I have described how it works in [When my knowledge base was connected directly to the model](https://docujai.com/node/159), and what it means to have your data layer in a format you own yourself, in [What one person can build in a week](https://docujai.com/da/node/225).

The point isn't that I did it first. The point is that when I wrote that my homemade note system was more AI-ready than most companies' CRMs, it was a contrarian claim. Now Google is releasing a specification that says exactly the same thing: the most important AI investment is not the model. It's whether your knowledge is in a format both parties can read.

## The three things management should take away

**The format trumps the platform.** The expensive data catalog, the closed wiki, the smart SaaS solution — they tie your knowledge to a vendor. Markdown in a version-controlled archive does not. When Google itself — which makes money on platforms — releases a *vendor-neutral* format, it's worth noticing. It's about who owns your data, where it's located, and whether the system can survive the vendor disappearing.

**Same file for human and machine.** The big saving is not technical. It's that you no longer have to maintain one version for employees and another for systems. The file a consultant reads in their editor is the same file the agent fetches context from. No translation, no double work, no versions drifting apart.

**AI readiness is a writing problem, not a purchasing problem.** The uncomfortable truth behind both OKF and my own experience: an AI agent is only as good as the knowledge you have written down and structured. SMEs run on tacit knowledge — what the experienced just *know*. That knowledge is invisible to a machine until someone sits down and makes it explicit. It's not a job you buy your way out of with a license. It's work that needs to be done before the technology can even help.

## What I would do in your company

You don't need IWE or Google's tools to get started. You need a place where your most important knowledge is written down, structured, and version-controlled — in a format you own yourself. Start with what a new employee or a new agent would need to know: what are our products, our processes, our customers, our rules. Write it down. Link it together.

It's not new, and it's not fancy. But it's the foundation for everything that comes after. Google has just given it a name. The principle is older: write down your knowledge so it can be used — by humans and machines at the same time.