Writing
On company knowledge and agentic infrastructure.
Thinking on the CIM category, ontology architecture, and what it takes to deploy agents that actually know how your company works.
A plain-language definition — what an ontology is, why it is not a database or a folder of documents, and why agents need one.
a16z is right that agents fail without context. But the context layer being described by most vendors is still too narrow. Here is what a properly structured company knowledge architecture actually requires.
Every major software category was made inevitable by a new form of coordination. Agentic AI has created the conditions for a new one. This is what we believe about it, and why we are building what we are building.
Retrieval-augmented generation solves a search problem. Enterprise agents have a knowledge problem. These are not the same thing, and conflating them is why 70% of RAG implementations fail to meet their objectives.
Palantir proved that enterprise ontology is the right architecture. They also proved it costs tens of millions to deploy. The thesis is correct. The delivery model is broken.
Every year, $31.5 billion in organizational knowledge evaporates. Not through security breaches or system failures — through normal employee turnover. This is not a people problem. It is an infrastructure problem.