The SQL ran fine. 58% of the answers were wrong (AI agents on a real warehouse)

The SQL ran successfully, yet 58 % of the answers were incorrect, highlighting AI agents’ limitations in real warehouses. To adopt it, first test the model on a