dbt Labs

Recently shipped dbt-core v1.12.0rc3 and v2.0.0-alpha.4, and the OSI spec became Apache Ossie (Incubating).

The v1.12 release candidate line progressed through three rounds. rc1 removed the experimental dbt login command and the bundled dbt-state plugin as a breaking change. rc2 set flags introduced in 1.9 and 1.10 to true by default, another breaking change. rc3 added CLI usage hints, fixed UDF dependencies being dropped from the DAG, and reduced parse-time memory by virtualizing macros. The direct_parents attribute for model nodes landed in the beta, and Fusion-specific warn_error_options names are now tolerated rather than erroring.

The v2.0 alpha gained deeper warehouse and catalog support. DuckDB can read and write Horizon and Unity (Iceberg REST) catalogs via duckdb 1.5.4, with catalog-aware materializations for DuckLake, Iceberg REST, and local filesystem catalogs. Redshift uses USE/RESET USE per node for cross-database operations and SHOW GRANTS ON TABLE for grants when datasharing is enabled. The --empty flag works with the seed command, and a Homebrew distribution (brew install dbt) landed. Versioned models now automatically create a latest-version pointer view, and the maximum seed size for state comparison is configurable.

The OSI open semantic layer spec entered the Apache Incubator under a new name. The project is now Apache Ossie (Incubating), renamed from Open Semantic Interchange to avoid confusion with the Open Source Initiative. Governance and long-term stewardship move to the Apache Software Foundation.

Context engineering emerged as a discipline. Three blog posts laid out a practical framework: structuring information for AI reasoning, building agentic data stacks on dbt, and migrating to dbt by starting fresh rather than lift-and-shifting legacy workflows.

The v1.11.x line got one more fix. v1.11.12 stopped emitting spurious deprecation warnings for state model configs like pre_clone and lag_tolerance.