Databricks details Lakebase branching for parallel coding agents

Databricks on October 8, 2026 published a workflow where every parallel coding agent and pull request runs against its own isolated, ephemeral Postgres database via copy-on-write branching in its Lakebase service. A branch is created in under a second regardless of database size, has its own compute, and scales to zero when idle.

• Lakebase branches are created in under a second regardless of database size
• Branching uses copy-on-write: storage is shared, billed only for changes
• Each branch has its own compute and scales to zero when idle
• Branches are not merged back; schema changes are promoted via migrations