Azure Databricks just shipped something that makes me wonder why this didn’t exist years ago. You can now create copy-on-write branches of your Lakebase production database with a single command, then point GitHub Copilot agent mode at it for debugging AI apps and agents. No copy of production data floating around in staging environments that nobody manages properly.
Here is the trick. It uses copy-on-write semantics, which means the branch starts as a metadata pointer to the same data blocks as production. No data is physically duplicated until you write to it. That makes branching near-instantaneous and practically free in storage cost. You can spin up a branch, let Copilot agent mode loose on it to debug some RAG pipeline, then delete the branch when you are done. Production data never leaves its cage.
The GitHub Copilot connection is the interesting part. Agent mode in Copilot can traverse schema, run queries, inspect data distributions, and test AI application logic directly against the branch. Since the branch sees a point-in-time snapshot of production, you get real-world data distributions without the risk of leaking PII or corrupting the source.