Do You Really Need an Ontology? We Measured It on a Small Coding Task and an Insurance Data ChatbotThanks to Palantir, "ontology" is everywhere these days. Does it actually help AI? We added and removed an ontology in two places, a small coding task (a Seoul bike-share fare module) and a chatbot that answers questions over a 13-table insurance database, and measured accuracy and cost. On the small task it only added cost. On the domain-heavy data, accuracy went from 62% to 82%.
Dante (Quokka Labs)Oct 29, 2026 
