Notes from the layer underneath.
Architecture decisions, post-mortems, and field notes from the multi-tenant agent infrastructure I build.
-
The Through-Line: Ontology as Operational Substrate
An ontology is not documentation - it is the runtime substrate an AI control system can stand on.
-
The Semantic Stack: RDF, OWL, and SHACL as Engineering Materials
RDF, OWL, and SHACL are engineering materials with real performance trade-offs, not academic formalisms.
-
Decentralized Org Control: First, a Decentralized Ontology
Decentralized control is downstream. Before an organization can be run by an evolving ontology instead of a CEO, you need a system that creates and maintains that ontology without a central owner - and that is the hard part.
-
Beyond 90% Accuracy: Building AI Agents That Survive Production
A 90%-accurate agent feels finished and isn't. Over twenty steps it's a coin flip you lose. The fix isn't a smarter model — it's a cautious architecture: measurable doubt, reversible actions, grounding, and memory that learns from every correction.
-
Building OmoiOS
How I built a self-hostable agent platform before background agents had a name, and what I learned shipping it without finding product-market fit.
-
Marketing is harder than building
Notes from a builder finding out that the easy part was the code.