BetaWeekly digests are a beta — we're trying something new. Feedback welcome.
dbt Core v2.0 rewrites the rules under Apache 2.0
July 20–26, 2026
dbt Labs announced dbt Core v2.0 under an Apache 2.0 license with a Rust-based Fusion engine, while retiring the Snowflake Native App for November 2026. Dagster added native dbt project orchestration on Snowflake, and ClickHouse published a series of practical tooling guides for observability migrations.
The week's single biggest story is that dbt Core has a new foundation. dbt Core v2.0 moves the open-source project to the Rust-based Fusion engine under an Apache 2.0 license, delivering dramatic CLI speed improvements and Parquet artifact files for local DuckDB queries. For teams running dbt at scale, this is a genuine step-change in developer experience — the modular architecture means the Fusion engine acts as a superset, with SQL comprehension features that dbt Core alone won't include. The announcement comes alongside news that the dbt Snowflake Native App will be retired in November 2026, giving users roughly 16 months to migrate workflows. The alpha release cycle continues with dbt-core v2.0.0-alpha.5, which fixes a minijinja mutable-list panic that cascaded into lock-poisoned panics and adds catalog_database support for Snowflake, BigQuery, and Databricks, plus defer_job_id support for job-specific manifest downloads.
Orchestration meets the new dbt
Dagster's 1.13.15 release introduces the SnowflakeDbtProjectComponent (Preview) for natively orchestrating dbt projects on Snowflake — a move that aligns with the growing trend of dbt-native orchestration layers. The release also includes documentation for the new Dagster MCP server, making it easier to integrate AI agents into pipeline workflows.
ClickHouse: observability migrations and agentic infrastructure
ClickHouse shipped a flurry of releases — v26.7.1, v26.6.2, v26.5.6, v26.4.5, and v26.3.17 — but the real story is in the tooling. The team published a detailed guide on migrating Datadog telemetry with the OpenTelemetry Collector, addressing a common pain point: Datadog agents and SDKs may already be deployed across thousands of hosts, and replacing all of that instrumentation before evaluating another backend creates a massive upfront project. A separate deep-dive on PostgresBench measures the impact of High Availability configurations on managed Postgres performance across vendors, following the community's strong reception to the original benchmark methodology. Meanwhile, agentic infrastructure with clickhousectl demonstrates a multi-region observability platform across three continents built entirely by a coding agent, and a benchmark of the ClickStack MCP server shows 18% more accurate root cause outcomes compared to generic SQL interfaces.
JSON ingestion and integer correctness
Tinybird's latest changelog corrects JSON integer ingestion behavior, addresses up to three mutations per alter in Classic, and adds three months of real-time Pipe statistics, plus experimental v1 ingestion in the CLI. It's a focused quality-of-life release for teams doing real-time analytics on streaming data.