dbt v2 goes GA and the analytics engine gets a Rust rewrite

dbt v2 ships as a Rust engine

The week's dominant story is dbt Labs' GA release of dbt v2 and dbt State, announced at dbt Summit 2026. dbt v2 is a full Rust rewrite that parses a 10,000-model project up to 10x faster than v1, distributed as the superset dbt and the Apache 2.0 subset dbt-oss. BigQuery, Databricks, DuckDB, Redshift, and Snowflake adapters are GA; ClickHouse and Spark remain in beta. Alongside it came Fivetran Context Layer in private beta with the open source Agents Schema standard, dbt Charts in public beta for version-controlled BI as YAML, and a set of Wizard experiences plus Lake Compute, a DuckDB-based single-node engine for Apache Iceberg tables. The summit keynote framing positions all of this as infrastructure for AI-driven analytics engineering — an engine fast enough to keep up with agents writing and rewriting models.

The practical payoff for teams is dbt State going GA across Snowflake, BigQuery, Databricks, and Redshift. Instead of a schedule, it inspects model SQL and warehouse metadata each run to decide whether to build, skip, clone, or defer a node — no selection syntax, no manifest scripts — and works with your own orchestrator, including Airflow, Dagster, GitHub Actions, or a laptop. Pricing is consumption-based on daily active target tables, with reuse after the first in a day free.

ClickHouse deepens its dbt integration

ClickHouse's own adapter for dbt v2 entered public beta, powered by the Rust engine, and ClickHouse joined the dbt platform in private beta for both open-source ClickHouse and ClickHouse Cloud. On the core side, dbt-core v2.0.1 renamed CLI branding from Fusion/dbt-core to dbt (proprietary) and dbt-oss, and the 2.0.0 release added agent skills installation gated behind a new ai_provider flag, native Databricks metric view materializations, and an adbc_clickhouse driver. The 2.0.5 patch fixed BigQuery incremental tables that could be written with REPLACE TABLE instead of MERGE.

Observability and orchestration converge on ClickHouse

ClickHouse's other announcements all point the same direction: bring metrics, logs, and traces under one query engine. The new TimeSeries Engine offers PromQL support as a drop-in Prometheus replacement, letting you query Prometheus metrics in ClickHouse Cloud without rewriting them in SQL. Replica-aware routing reached public beta, pinning requests to a single replica so temporary tables and named sessions stay visible over HTTP or the native protocol. The official ClickHouse provider for Apache Airflow landed with standard SQL operators, bulk inserts, and shared setup across self-managed Airflow and Astronomer, while ClickStack's August update added dashboard variables, chart formulas, PromQL support, and LLM observability. Orchestration also moved at Dagster, which added deployment capacity alert policies in Dagster+.

Agent ergonomics got attention too: Tinybird's MCP now lets you choose MCP response formats and inspect query plans, accepting CSVWithNames, JSON, or Markdown and supporting explain=true before executing a query. Rounding out the week were routine ClickHouse stable and LTS builds and a handful of low-impact dbt-core patches, including v1.12.5 fixing the Dimension protocol and bumping the metricflow floor.