dbt vs Lakeflow Declarative Pipelines on Databricks: which should you use?

On Databricks, teams building transformation pipelines usually weigh two options: dbt, the widely used SQL modelling framework, and Lakeflow Declarative Pipelines (formerly Delta Live Tables), Databricks’ native declarative pipeline engine. They overlap, but they are strongest at different layers.

The short answer

• Use Lakeflow Declarative Pipelines for ingestion, streaming, change data capture and slowly changing dimensions, especially close to the sources.
• Use dbt for SQL business modelling, especially if your analytics engineers already know dbt or you want logic that can move between platforms.
• Use both in many cases: Declarative Pipelines for bronze and silver, dbt for gold.

How they compare

	dbt	Lakeflow Declarative Pipelines
What it is	SQL (and Python) modelling framework	Databricks-native declarative pipeline engine
Languages	SQL with Jinja; Python models on some platforms	SQL and Python
Batch modelling	Excellent	Good (materialized views)
Streaming	Limited	Native streaming tables
Change data capture, SCD	Incremental models and snapshots, written by hand	AUTO CDC with SCD type 1 and 2 built in
Data quality	Tests (unique, not null, relationships, custom)	Expectations that warn, drop or fail per row
Lineage and docs	dbt docs and lineage graph	Unity Catalog lineage, pipeline graph
Orchestration	dbt Cloud, Lakeflow Jobs, Airflow	Managed by the pipeline; scheduled with Lakeflow Jobs
Compute	Runs SQL on a Databricks SQL warehouse	Managed pipeline compute (serverless available)
Portability	Runs on many platforms	Databricks
Community	Very large, many packages	Databricks ecosystem

What usually decides it

1. Streaming and CDC needs. If sources change continuously, Declarative Pipelines save a lot of hand-written logic.
2. Team skills. Analytics engineers fluent in dbt are productive immediately; data engineers often prefer Declarative Pipelines for ingestion.
3. Multi-platform strategy. dbt keeps modelling portable if another warehouse is in the picture.
4. Data quality style. Row-level expectations in the pipeline versus tests after the fact. Both work; many teams use both.

The short version

Declarative Pipelines for ingestion, streaming and CDC; dbt for portable SQL business modelling; both together is a sound and common architecture on Databricks.

See how they fit in our migration guides, or compare platforms with Databricks vs Microsoft Fabric. Our data platform engineering service designs pipelines like these.