Databricks Field Guide

Reference · Chapter 30

Delta Live Tables, Mosaic AI and other Databricks renames

Reviewed by TechFabric engineers. How we write this guide →

Delta Live Tables (DLT) is now called Lakeflow pipelines, and its documentation lives under docs.databricks.com/aws/en/ldp/, with a page titled What happened to DLT? explaining the change. Existing DLT code keeps running without a migration. The same is true of the other renames below: Asset Bundles, Vector Search, Unity AI Gateway, Genie spaces, Lakeview and Workflows all have new names, and the old ones survive in code, API paths and billing data.

That's why a new engineer can follow a DLT tutorial, fail to find the product in the current docs, and decide it was retired, and why a finance analyst can ask what "DLT" is doing on an invoice for a platform that doesn't sell anything by that name. Both are the same problem, and the table below is the fix we'd hand to either of them.

The short version

Most of what you learned under an old name still works under the new one. Delta Live Tables is now Lakeflow pipelines, Databricks Asset Bundles are Declarative Automation Bundles, Vector Search is Databricks AI Search, Genie spaces are Genie Agents, Unity AI Gateway is Unity Gateway, and the gateway settings on a single serving endpoint are now "AI Gateway (legacy)". Running code keeps working in every case. Search, onboarding and cost reporting are what break.

Which Databricks products were renamed? #

Each row was checked on 23 October 2026 against the Databricks page that uses the new name; the last column gives that page's own "last updated" date.

You'll still see The current name Where the old name survives Databricks page, last updated
Delta Live Tables, DLT Lakeflow pipelines, built on Apache Spark Declarative Pipelines (in Spark from 4.1) import dlt, classic SKUs that start with DLT, billing_origin_product = 'DLT', event log schemas with dlt in the name What happened to DLT?, 11 September 2026
Databricks Asset Bundles, DABs Declarative Automation Bundles The databricks bundle CLI commands and the databricks.yml file are unchanged What are Declarative Automation Bundles?, 11 September 2026
Mosaic AI Vector Search, Databricks Vector Search Databricks AI Search Older links under /vector-search/, VectorSearchClient in existing Python code (the current SDK is databricks.ai_search with AISearchClient), the vectorsearchendpoints REST reference and billing_origin_product = 'VECTOR_SEARCH' Databricks AI Search, 21 October 2026
Unity AI Gateway Unity Gateway ai-gateway in API paths and in the OAuth scope apps must request, and the older ucode CLI command beside the new ug Unity Gateway overview, developer site, no date shown
Mosaic AI Gateway, AI Gateway on a serving endpoint AI Gateway (legacy), with Unity Gateway as the generally available replacement Endpoint-level gateway settings and their inference tables Governance for model serving endpoints (legacy), 8 October 2026
Genie spaces, AI/BI Genie Genie Agents, with Genie One as the business user's front door Older tutorials, and anything that says "space" Genie Agents, 7 October 2026
Lakeview dashboards AI/BI dashboards The REST API is still /api/2.0/lakeview/dashboards AI/BI dashboards API reference, no date shown
Databricks Workflows Lakeflow Jobs Nothing in code: the API, the bundle key and the billing product were always jobs Lakeflow Jobs, 9 October 2026

The Mosaic prefix is the one we'd warn new readers about. It came with the MosaicML acquisition and spread across the AI products, so a great deal of writing from 2024 and 2025 says Mosaic AI Vector Search, Mosaic AI Model Serving or Mosaic AI Agent Framework. The current pages have dropped it: the AI Search page opens with "Databricks AI Search (formerly Databricks Vector Search)" and doesn't use the word Mosaic anywhere. In an architecture document, read Mosaic as a date stamp.

One more change sits beside the renames without being one. Agent Bricks Knowledge Assistant and Supervisor Agent are now legacy offerings: Databricks points new low-code agents at Genie Agents and code-based agents at the Agent Bricks CLI. Agents covers what to build instead.

Databricks renames a product

Docs and UI
use the new name

Code and API paths
keep the old one

Billing and system tables
keep the old one

Tutorials and posts
from before the rename

Change at the next edit

Fix cost reports now

Fix onboarding pages now

Where an old Databricks name keeps turning up after a rename

How do I move DLT code to the new pipelines API? #

Databricks' position is that no migration is required, and we agree with it as far as running code goes. Existing dlt pipelines keep working. We'd still move code to the new module the next time someone touches the file, for one practical reason: the new names line up with Apache Spark Declarative Pipelines, so the parts of the pipeline that use only open source features will also run outside Databricks.

The mapping from the Lakeflow pipelines Python reference, last updated 19 October 2026:

What it does DLT syntax Current syntax In Apache Spark too?
Import import dlt from pyspark import pipelines as dp Yes
Streaming table @dlt.table over a streaming DataFrame @dp.table Yes
Materialized view @dlt.table over a batch DataFrame @dp.materialized_view Yes
View @dlt.view @dp.temporary_view Yes
Change data capture dlt.apply_changes(...) dp.create_auto_cdc_flow(...) No
Expectations @dlt.expect(...) @dp.expect(...) No

The second and third rows are the change that catches people. Under DLT one decorator, @dlt.table, produced either a streaming table or a materialized view depending on what the function returned. The new API makes you say which you meant:

# Before: one decorator, two behaviours
import dlt

@dlt.table
def orders_raw():
    return spark.readStream.table("bronze.orders")      # became a streaming table

@dlt.table
def orders_daily():
    return spark.read.table("silver.orders").groupBy("order_date").count()   # became a materialized view

# After: the intent is in the decorator
from pyspark import pipelines as dp

@dp.table
def orders_raw():
    return spark.readStream.table("bronze.orders")

@dp.materialized_view
def orders_daily():
    return spark.read.table("silver.orders").groupBy("order_date").count()

That explicitness is the reason we'd do the change deliberately rather than with find and replace. A mechanical swap of dlt for dp leaves @dp.table on a batch function, and a reviewer reading it afterwards can no longer tell whether a materialized view was intended. The last three rows matter for portability: AUTO CDC and expectations exist only in Lakeflow pipelines, so a pipeline that uses them stays on Databricks whatever its imports say.

Why does the Databricks bill still say DLT? #

This is the rename that reaches finance. The classic SKUs for Lakeflow pipelines still begin with DLT, and the billing system table reference (last updated 18 October 2026) still lists DLT as the billing_origin_product for "Lakeflow pipelines workloads", beside JOBS for Lakeflow Jobs, SQL for Databricks SQL and VECTOR_SEARCH for AI Search. A cost report that groups by product therefore shows a line nobody can find in the current product list, and in our view the fix belongs in the report, not in an email explaining it each month:

-- Spend by SKU over the last 30 days, labelled with today's product names
SELECT
  CASE billing_origin_product
    WHEN 'DLT'  THEN 'Lakeflow pipelines'
    WHEN 'JOBS' THEN 'Lakeflow Jobs'
    WHEN 'VECTOR_SEARCH' THEN 'Databricks AI Search'
    ELSE billing_origin_product
  END                       AS product,
  sku_name,
  SUM(usage_quantity)       AS dbus
FROM system.billing.usage
WHERE usage_date >= current_date() - INTERVAL 30 DAYS
GROUP BY ALL
ORDER BY dbus DESC;

We'd keep that CASE in one view that every cost dashboard reads, so the translation happens once. What It Actually Costs and Cost and Performance cover what to do with the numbers once they're labelled.

Keeping internal docs from aging #

Two habits do more than watching release notes. When we write anything internal that should outlive the quarter, we'd name the capability beside the product, because "the declarative pipeline framework" survives a rename and "DLT" doesn't. And when a name you know stops returning results, search the docs for the old name with "formerly". Not every renamed page mentions its old name (the Lakeflow Jobs overview never says Workflows), so the Glossary carries both names for every renamed product in this guide.