Building Knowledge Infrastructure That Actually Travels With Your Workflow

Key takeaways
✦ Google just formalized a pattern most AI builders have been solving informally: how agents find and use organizational knowledge
✦ The problem isn't the model. It's context assembly: scattered docs, siloed wikis, no portable format
✦ Domain primitives built on OKF become durable infrastructure, not one-off prompt hacks

Most AI agents fail before they start.

Not because the model is wrong. Because the knowledge it needs is scattered across a nested shared drive, a Confluence page, a shared wiki, or an important KPI framework someone emailed a while back.

What Google just shipped

On June 12, Google Cloud published the Open Knowledge Format (OKF) v0.1: an open spec for packaging organizational knowledge so agents can actually use it. Plain markdown files. YAML frontmatter. No proprietary SDK required.

The core insight: the context assembly problem is the same one every agent builder solves from scratch. OKF bets the fix is a format, not another platform.

The problem it surfaces

I've been building what I call domain primitives: structured knowledge stores encoding not just data, but the decisions, definitions, and business logic surrounding it. Agents need that metadata. Raw data alone isn't enough.

OKF gives that work portable architecture. It formalizes a pattern emerging informally across Obsidian vaults, agent markdown repos, and llms.txt files, pinning down the conventions that let knowledge written by one producer be consumed by a different agent without translation.

What I built on top of it

I built an ingestion agent that takes any file format and produces an OKF-structured domain primitive. Drop in a PDF brief, a CSV, a strategy doc, a campaign taxonomy. The agent extracts the concepts, structures them as OKF bundles, and outputs something any downstream agent can consume as grounded context. A knowledge layer that travels with the workflow.

What most orgs miss

The context assembly problem doesn't get solved by a better model. It gets solved by unglamorous knowledge engineering done before the agent runs. OKF doesn't do that work for you. It gives the work a home.

The organizations pulling ahead aren't prompting harder. They're building knowledge infrastructure agents can actually navigate.

That's the plumbing. And the bill is coming due for every org that skipped it.

This is a pretty clean demonstration of how I typically work: convert frontier AI concepts into organizational leverage before they become conventional wisdom. If your team is wrestling with the context assembly problem, reach out.

Link to the OKF repo: https://github.com/GoogleCloudPlatform/knowledge-catalog/tree/main/okf