|
In the last seven days: eight shipped, eight in preview, five with a date on it and six new concepts on Lakenaut.
21 changes for you: one breaks something, five needs action, ten worth evaluating and five nice to know.
 |
New features |
|
-
|
|
Databricks announced general availability of AI Runtime, its managed serverless GPU compute for training and fine-tuning models next to Unity Catalog data. |
-
|
|
A Genie Code task runs a prompt autonomously as a step in a job, can read upstream task outputs and call tools, and returns a link to the resulting Genie Code chat. |
-
|
|
In the 19 October 2026 Databricks Runtime 19 update, Spark Connect DataFrame execution metrics no longer show internal operators and metrics tied to row filters and column masks. |
-
|
|
A custom identity claim can now be added to a service principal's OAuth token and used to filter rows in a dynamic view or an ABAC policy. |
-
|
|
The Python row-based transformWithState API can now run state operations and user logic concurrently, which the release note says raises throughput with only minor code changes. |
-
|
|
They record the pipelines and pipeline updates in an account, and are documented in the jobs system table reference. |
-
|
|
Saving a Lakeflow job without changing its settings no longer adds rows to system.lakeflow.jobs or system.lakeflow.job_tasks, and change_time stays at the last real change. |
-
|
|
A new DeepSeek V4.1 Flash version, labelled 1017, can now be called on Unity Gateway through Foundation Model APIs and accepts text and images. |
 |
Previews and beta |
|
-
|
|
Lakebase Insights (Beta, formerly Lakebase Advanced Postgres Telemetry) is now set up in each project's Settings by users with CAN MANAGE, with on-demand runs. Breaking: If you scheduled Insights through an observability configuration, enable Insights and set an analysis frequency in each project's Settings… |
-
|
|
df.cache() and df.persist() now work on serverless compute, so an expensive result can be reused for the rest of a session. |
-
|
|
It adds multipart uploads, so files larger than 5 GB can be uploaded, and conditional operations, which upload, download or delete a file only if its current version matches. |
-
|
|
Serverless notebook and Python script tasks in Lakeflow Jobs can now take their job environment from a pyproject.toml file, set in the Jobs UI, the Jobs API or a bundle. |
-
|
|
Shared Serverless for Asset Bundles lets notebooks and files in the same bundle share one serverless compute and an environment defined in the bundle-root pyproject.toml. |
-
|
|
It screens MCP tool calls and their results for jailbreak and prompt injection attempts, and can be applied on input, output, or both. |
-
|
|
It ingests enterprise audit events from Box into Databricks. |
-
|
|
It ingests users, groups, applications, service principals, directory roles, identity configuration and audit and sign-in logs. |
 |
Deprecations and end of support |
|
-
|
|
DeepSeek V4.1 Flash is the recommended replacement. |
-
|
|
Anything still relying on it needs a plan before then. |
-
|
|
Anything still relying on it needs a plan before then. |
-
|
|
Anything still relying on it needs a plan before then. |
-
|
|
The release note names DeepSeek V4.1 Flash (1017) as the recommended replacement and points to the deprecated and retired models page for migration guidance. Action required, by 19 Nov 2026: Move workloads that call DeepSeek V4.1 Flash to DeepSeek V4.1 Flash (1017). |
 |
New and rewritten concepts |
|
-
|
|
Databricks separates a vendor-managed control plane from a compute plane that processes the data, with storage in the customer's cloud, Delta Lake as the format. |
-
|
|
The Conversation API, embedding, Slack and Teams, MCP and the Supervisor Agent let a Genie Agent answer from anywhere, always as an identity whose permissions apply. |
-
|
|
Both pipeline interfaces build the same dataflow graph, so most of the choice is taste. The asymmetries are few and they all run one way: Python covers the whole feature set. |
-
|
|
Unity Catalog privileges are granted to users, groups and service principals and inherit from the catalog down. USE CATALOG and USE SCHEMA are the entry door. |
-
|
|
The human half of the Genie Ontology. Domains group assets by business purpose, Pages define what a business term actually means, and Genie prefers both over what it infers. |
-
|
|
BEGIN ... END compound blocks with local variables, loops, EXECUTE IMMEDIATE, condition handlers and cursors, and how to persist a working script as a Unity Catalog procedure. |
|