Oct 6 — Oct 13, 2026
This week was defined by major investments in operational and agentic developer infrastructure, led by the rollout of Lakebase serverless Postgres and the GA of user-delegated authorization in Databricks Apps. Zero-copy enterprise federation and unified agent management tooling across the SDKs also saw substantial momentum.
1.Lakebase Postgres rolls out as the transactional and state backend for agentic apps
Databricks heavily showcased Lakebase, a managed serverless PostgreSQL database built directly into the platform. Coverage focused on using Lakebase for sub-second copy-on-write database branching for autonomous coding agents, integrating with Replit and Genie App Builder, and leveraging Spark-driven bulk ingestion to load terabytes of operational data in minutes.
Sources
- Building on Lakebase | Episode 2: Anatomy of an AI app: build the operational + agentic appVideo · Databricks Skill Builder · Oct 13
- Building an Expense Management App with the Genie App Builder Leveraging Lakebase + AI for Image ScanningCommunity · reddit · Oct 13
- Lakebase and Agentic SDLC: Branching Databases for Coding AgentsNews · databricks-blog · Oct 8
- Load terabytes of data in minutes into Lakebase PostgresNews · databricks-blog · Oct 8
- How to build governed enterprise apps on Databricks with Replit and LakebaseNews · databricks-blog · Oct 8
2.Agent Bricks and SDK-level AgentKit formalize the AI agent runtime stack
Databricks introduced Agent Bricks to provide secure execution sandboxes, model routing, and agent hosting, while officially rolling AgentKit into the Python, Go, and Java SDKs to replace the deprecated Mason service. Accompanying MLflow 3.17 releases added managed GenAI review queues and improved OpenTelemetry trace rendering for coding agents.
3.Databricks Apps secures GA on-behalf-of-user auth and Model Serving securables
Databricks Apps reached general availability for on-behalf-of-user (OBO) authorization, enabling applications to run under end-user permissions rather than fixed service accounts. In parallel, platform SDKs and MLflow added support for Unity Catalog model service endpoints as first-class securable resources within application manifests.
4.Zero-copy Lakehouse Federation pushes deeper into enterprise ERP and HCM data
Practitioner discussions increasingly favored Unity Catalog Query Federation over traditional ingestion pipelines to curb ETL sprawl. Databricks reinforced this direction by launching the Workday Data Connect federation connector over shared Iceberg tables and detailing zero-copy integration architectures for SAP data products.
Sources
- Federation vs. ETL: Why I’m starting to hate moving data just to look at it.Community · reddit · Oct 13
- Stop migrating data to Databricks: Are you using Lakehouse Federation yet?Community · reddit · Oct 13
- Announcing Workday Data Connect federation in Unity CatalogNews · databricks-blog · Oct 8
- Databricks and SAP: Building the Next Generation of Enterprise Data and AnalyticsVideo · Databricks Skill Builder · Oct 8