Fabric Dataflows Gen2: Top New Features

Key insights

• Dataflows Gen2 is Microsoft Fabric’s next-gen ETL engine for preparing data into Lakehouse, Warehouse, and SQL targets.
  It fits both data engineers and business users who need scalable, governed data pipelines.
• Use Preview-only steps to speed authoring: they show design-time results without affecting refresh runs.
  This reduces wait time while you refine transformations.
• Turn on the Modern Query Evaluation Engine (under Scale) and enable Partitioned Compute to get faster, parallel execution and lower runtime latency.
  These changes often cut compute time for large refreshes.
• The new two-tier pricing model bills higher (12 CU) for the first 10 minutes and lower after that, so short bursts cost more per minute than long runs.
  Combine incremental runs, partitioning, and the modern engine to reduce overall CU spend.
• Copilot assists authoring by suggesting calculated columns and offering an “Explain step” feature to clarify transformations.
  The Modern Get Data experience and schema-aware navigation make discovering and shaping sources easier.
• Dataflows Gen2 now writes to more targets (Lakehouse, Warehouse, SQL, ADLS and a Snowflake preview) and integrates with Fabric pipelines, variable libraries, and parameters for CI/CD and automation.
  Enable features where noted in settings, choose partitioning for parallel runs, and use incremental refresh for Lakehouse tables to scale ETL efficiently.

Pragmatic Works released a detailed YouTube walkthrough that summarizes recent updates to Fabric Dataflows Gen2. The video highlights improvements in authoring speed, query execution, pricing, and destination support, and it shows practical steps to enable new features. Consequently, the update aims to make ETL processes faster and more cost-efficient for both data engineers and analysts.

What the update adds and who benefits

The presenter explains that Fabric Dataflows Gen2 now targets teams that need scalable, governed ETL across Fabric destinations. For example, direct writes to Lakehouse tables and Warehouses reduce handoffs between services, which improves operational flow. Moreover, the update adds preview connectors and schema-aware navigation that help analysts discover and reuse data quickly.

At the same time, Pragmatic Works points out that both citizen developers and professional engineers will gain from the improvements. Thus, small teams can prototype faster while large teams can enforce governance and CI/CD patterns. In short, the release balances ease of use with enterprise needs.

Performance improvements and cost tradeoffs

One major change is the introduction of the Modern Query Evaluation Engine and Partitioned Compute, which together speed up query execution and enable parallel processing. As a result, refreshes run faster and heavy transformations can complete in less time. However, faster runs may use more resources during peak execution, which means teams must weigh speed against cost.

Pragmatic Works also explains the new two-tier pricing model and how it affects compute cost. The model charges higher CU rates for the first ten minutes and lower rates for longer runs, which reduces the unit cost for lengthy batch jobs. Therefore, architects must decide whether to restructure workloads for long continuous runs or to optimize for many short runs, because each approach has different cost and operational implications.

Authoring experience: previews, Copilot, and schema navigation

The video highlights Preview-only steps, a feature that speeds up design time by letting authors inspect transformations without triggering full refreshes. Consequently, developers can iterate quickly and avoid long waits while testing logic. This approach reduces development friction but requires careful validation before production runs, since preview data may not reflect full refresh behavior.

Additionally, the presenters demo how Copilot assists with calculated columns and an "Explain step" feature, which helps users understand transformation logic. While Copilot can accelerate authoring, the video cautions that AI suggestions should be reviewed, because automatic transformations may not match governance rules or performance best practices. Thus, teams should use Copilot as a productivity aid, not as a substitute for review.

Expanded destinations and schema-level governance

Another important update is the broader set of destinations, including Lakehouse tables, Fabric Warehouse, and a preview connector for Snowflake. With these targets, Dataflows Gen2 supports more hybrid architectures and reduces the need for intermediate export steps. Moreover, schema-aware navigation and database schema controls improve discoverability and make it easier to apply naming and access policies.

On the other hand, supporting multiple destinations introduces extra governance and testing work. Teams must ensure consistent schemas and data quality across different sinks, and they should adopt automated tests to avoid drift. Therefore, while flexibility increases, operational complexity also rises and demands stronger release practices.

Operational considerations and next steps

Pragmatic Works demonstrates using variable libraries and parameterized dataflows to enable CI/CD and environment-specific behavior. As a result, teams can move dataflows through development, test, and production with fewer manual changes. Nevertheless, this adds configuration complexity that requires good documentation and disciplined change control.

Finally, the presenters suggest practical ways to lower CU spend, such as enabling incremental refresh for Lakehouse tables and choosing partition strategies carefully. While these techniques reduce compute cost, they require planning around data change patterns and retention. Overall, the video provides useful guidance for teams preparing to adopt or scale Fabric Dataflows Gen2, while reminding viewers to balance performance, cost, and governance.

Keywords

Fabric Dataflows Gen2, Power BI Dataflows Gen2, Microsoft Fabric dataflows, Dataflows Gen2 features, Fabric Dataflows updates, Dataflows Gen2 performance improvements, Dataflows Gen2 lakehouse integration, Fabric data governance updates