Databricks publishes Temporal and Lakebase Postgres reference design for durable AI agents

Databricks has published a reference implementation that pairs Temporal, a workflow orchestration system, with Lakebase Postgres to make cloud AI agents durable.

According to Databricks' announcement on X, the design lets recorded work survive when a worker, container, or deployment is replaced, and keeps evidence and review state queryable so human decisions can arrive days later.

Databricks says Unity Catalog remains the governed policy source, with policy reaching the agent system through synced tables. The report is a single announcement from Databricks; no independent evaluation of the design was reported.