Charts built for Chat
summary
dbt Labs is open sourcing dbt Charts, a declarative YAML-based language for dashboards to solve the friction between messy AI-generated code and restrictive BI tools. By moving chart definitions into code, the system enables AI agents to create governed, auditable, and version-controlled visualizations. This approach unbundles the visualization layer from the BI tool, allowing charts to be managed in Git repositories alongside data models. The goal is to create an open standard for the chart layer of the data stack that works seamlessly for both humans and AI.
key points
- dbt Charts uses a structured YAML language to define how data is visualized, while SQL remains the language for defining what data is to be seen. This allows for a complete interactive dashboard to be declared in a single, auditable file that can be rendered to various formats like SVG, HTML, or PDF.
- Integrating dbt Charts into a dbt project allows dashboards to live in the same Git repository as data models, enabling changes to be shipped through a single CI run and preventing breaking changes from reaching production through strict validation.
- The system is specifically designed for AI agents, providing a tight feedback loop through strict YAML and SQL validation and visualization checks that flag layout or data density issues before rendering.
who should read it
Data engineers and analytics engineers who want better version control for their dashboards and data teams using AI agents to automate report generation.
discussion positive
consensus
Most commenters agree that 'unbundling BI' and moving toward dashboards-as-code is a positive trend, especially for AI agents that require structured, auditable artifacts rather than free-form output. While there is general support for the open-source approach, some argue that YAML is an insufficient DSL for complex logic and that AI will eventually outpace the need for specific specs by generating primitive libraries directly. Despite these technical disagreements, the sentiment is largely supportive of tools that increase transparency and maintainability in data visualization.
top insight
A critical tension exists between using a structured data format like YAML for consistency and using a true programming language for the feedback loop; the latter allows LLMs to iterate via compile-time or runtime errors, which is far more effective for complex logic than guessing a schema. This perspective suggests that the 'bitter lesson' of AI might favor flexible, primitive-based generation over the strict specifications currently being proposed.
Generated by an LLM from the article and thread. Check the sources before relying on it.