Ishita Daga identifies three structural problems in enterprise data agents: ambiguity about authoritative sources, stale business context and different user preferences for apparently identical metrics. Larger models or longer prompts do not automatically settle those questions.
Her proposed information hierarchy starts with curated semantic definitions, then parameterised canonical queries, and finally more flexible database graphs. She recommends connecting actively maintained sources with logged corrections and regular evaluation so context changes become part of a repeatable maintenance process.
Team preferences remain an open problem in her account. Two teams can calculate a milestone interval differently and both be correct under their own definitions. Semantic layers can preserve alternatives and memory can retain preferences, but neither by itself guarantees routing a requester to the intended metric.
Watch on YouTube
