A city wants resident analysts using AI/BI Genie to discover the correct 311 domain tables when they ask natural-language questions about ticket metrics. What should data engineers configure?

AI/BI Genie instructions give domain context so natural-language questions resolve to the right Unity Catalog tables. For city 311 ticket metrics, engineers should document which gold or curated tables hold ticket volumes, statuses, and domains, plus preferred phrasings and join keys. Those instructions act as a front-desk briefing for Genie: resident analysts ask in plain language, and the assistant is steered toward the correct 311 objects instead of arbitrary bronze landings.

Disabling natural language and forcing memorization of three-level UC names rejects the Genie use case the question asks for. Replacing Unity Catalog with a shared Google Sheet of names pasted into each chat creates an ungoverned, stale side channel and does not configure Genie. Granting every Genie user unrestricted MODIFY on all bronze landing tables is a privilege anti-pattern: discovery guidance is not the same as write access to raw landings, and it risks corrupting municipal ingest.

Exam caveat: Genie instructions improve discovery and prompting quality but do not replace Unity Catalog privileges—users still need SELECT (and related) rights on the tables Genie surfaces. Instructions also need maintenance when 311 schemas change. Operational check: add Genie instructions naming the 311 metric tables, ask a sample question about open tickets by district, confirm Genie cites the intended gold tables, and verify parks or unrelated groups without SELECT still cannot retrieve rows.