The v2.0 alpha landed on June 1 alongside Snowflake Summit. The headline capability is deferral with user-defined functions — models that depend on UDFs can now defer to production references during development without resolving the full DAG. It's a meaningful workflow improvement for large dbt projects, though the alpha tag means real bugs will bite before this stabilises. Also shipping alongside the announcement: a dbt-confluent adapter bringing dbt support to Confluent Cloud for Apache Flink. The two together suggest dbt is pushing hard into streaming and non-warehouse runtimes, not just the batch SQL layer it built its reputation on.