project
Databricks governance layer cited as the boundary that inference and its outputs remain inside when AI Functions are used.
No evidence-backed relationships are recorded.
Its SQL scripting now maps cursors, temp tables and exception handlers with near-identical syntax. The piece it calls decisive, multi-table rollback, is where the published walkthrough runs out.
Reality
A dev.to walkthrough maps Git branches to Asset Bundle targets and stacks seven automated gates before a human approves prod. The mapping, not the YAML, is the load-bearing part.
Reality
An August 3 release note dropped the usage-privilege requirement for MANAGE, and it applies to grants already in your metastore. Nobody had to run a GRANT for the permission to start working.
Reality
Databricks is pitching SQL-callable AI functions as convenience. The consequential change is that predictions stop leaving the catalog, and the audit trail stops being a promise.
Reality