dbt Labs open sources declarative YAML language for agent-built dashboards

The event is the open-sourcing of dbt Charts, a declarative YAML language for defining dashboards. This moves dashboard creation from BI tool interfaces into code, specifically structured YAML files. The change targets the friction between agent-generated dashboards and governance, where agents currently produce messy code that is difficult to audit and maintain.

For engineers, adopting dbt Charts means learning a new YAML dialect and toolchain. The language wraps SQL queries in YAML to define charts, layouts, and variables, with Markdown for prose and Jinja for templating. While the language is designed to be readable, it introduces a new dependency and workflow. The CLI renders board files to static formats or serves them as a site, replacing BI tool export functionality.

The language is designed for extensibility and theming. Charts inherit styles from boards, boards from themes, and themes can be switched with a single line. This reduces repetition but requires engineers to structure their dashboards to leverage these cascading styles. The integration with dbt projects is deep, allowing charts to live alongside data models in the same Git repository, which simplifies version control and change tracking.

The trade-off is between the freedom of agent-generated code and the governance of structured YAML. BI tools currently offer narrow governance through their UIs, while raw code offers freedom but poor auditability. dbt Charts aims to provide both, but its effectiveness depends on how well engineers adopt the new language and toolchain. The material does not specify limitations, but the language's complexity (over 1,100 config options) suggests a learning curve and potential edge cases.