The failure is caused by inadequate differentiation between the responsibilities of the agent ' s structured SQL retrieval tool and its unstructured vector search tool.
The live inventory database is the authoritative source for current stock quantities and availability.
The product catalog vector index is intended for descriptive information, including product specifications, categories, and general product characteristics.
Although both tools function correctly, the agent repeatedly selects the catalog tool for availability questions, indicating that its tool-selection instructions do not sufficiently distinguish current operational data from static descriptive information.
Option B addresses this routing ambiguity directly.
The engineer should define the SQL tool as the required mechanism for questions involving current inventory levels, stock status, warehouse availability, and replenishment quantities.
The vector search tool should be explicitly limited to product descriptions, technical specifications, and catalog-related questions.
Option A identifies a possible limitation of the static catalog but does not explain the incorrect tool choice. Option C assumes a specific omission in the inventory tool description that the scenario does not establish. Option D identifies a related grounding weakness, but the primary observed failure occurs during tool selection.
Databricks agent frameworks support separate tools for structured and unstructured retrieval.
Explicit routing instructions and source-of-truth requirements should be evaluated through representative inventory and catalog queries.
Official Databricks reference: MCPs and agent tools and MCPs.
────────────────────────────────────────────────────
Submit