One thing I keep seeing in AI projects is that the problem is not always the model. Very often, the model is fine. The real problem is that the knowledge around it is messy.
Companies have documents in folders, notes in emails, rules in people's heads, processes in Excel, and then they expect an AI agent to understand everything perfectly.
It doesn't work like that.
This is why I find OKF — Open Knowledge Format interesting. It is a simple way to write knowledge in markdown files, with a small structured part on top, so both humans and AI systems can understand it better.
A simple example could look like this:
--- type: Business Process title: New Client Onboarding description: The process followed when a new client starts working with the company. tags: [client, onboarding, operations] timestamp: 2026-06-29 --- # New Client Onboarding When a new client signs, the operations team creates a client folder, collects the required documents, checks missing information, and assigns an internal owner. ## Important Rules - No project starts without signed approval. - Missing documents must be flagged before the first meeting. - The final client profile must be reviewed by a human. ## Related Knowledge - [Client Documents](./client-documents.md) - [Approval Workflow](./approval-workflow.md)
This is not impressive because it looks advanced. It is interesting because it is clear.
A person can read it, a developer can version it, and an AI agent can use it as context.
For me, this is where many companies need to focus before they run after the next "smart agent". AI is much better when the knowledge around it has some structure. Not perfect structure, because real business is never perfect, but at least enough to know what is true, what is related, and what needs human approval.
OKF will not magically fix bad data or bad processes. But it points to something important: the future of AI is not only bigger models. It is also better organized knowledge. And that part is not very glamorous, but it is where serious AI systems start.
Originally published on LinkedIn.