Fivetran and dbt Labs are rolling out a major update to their open-source transformation tool, putting to rest community fears that the project would go private. dbt Core v2.0, now licensed under Apache 2.0, is built on the same Rust-based engine as the company’s Fusion product, delivering noticeably faster command-line performance. The release also introduces dbt State, a caching layer that skips unchanged models during runs, and dbt Wizard, a coding agent tuned specifically for analytics engineering tasks.

The company says dbt State cuts customers’ compute bills by at least 30% by tracking changes at the column, table, and code level. If nothing has changed and the data is still fresh, dbt simply does not rebuild the model. Fivetran’s own large revenue model dropped from a 40-minute run to about 40 seconds after switching it on. The feature works with dbt Core plus Airflow and does not require the full dbt platform, though it is a paid add-on. dbt Wizard, meanwhile, plugs into any large language model and handles tasks like refactoring and autocorrect more efficiently than generic coding agents because it is purpose-built for analytics workflows.

The update comes as the company argues that the data stack must be rebuilt for an era where AI agents, not just humans, consume data. Fivetran and dbt Labs point to three structural problems at agent scale: broken trust from incomplete data, stalled scale from ungoverned data, and exploding costs from repeated retrievals. Their answer is an architecture called Open Data Infrastructure, which relies on open formats like Iceberg and Delta, separates storage from compute, and provides rich metadata for governance. Early adopters include Zendesk, Shutterstock, and Inova Health, the latter of which compressed a four-year modernization plan into six months.

Source: dbt Labs Blog