Flint: AI Agents Get Smarter Charts

Microsoft Research has unveiled Flint, a new visualization language designed to simplify AI-driven chart creation. This intermediate language allows AI agents to produce sophisticated, visually polished charts from concise, human-editable specifications, addressing a significant pain point in current AI visualization workflows.

Traditional visualization libraries often force a trade-off: either simple specs yield uninspired charts, or polished visuals demand verbose, error-prone configurations. This challenge intensifies as large language models (LLMs) and AI agents increasingly handle visualization tasks, often struggling with complex, low-level specification details.

Flint streamlines this process by leveraging semantic data types. These types guide the compiler in selecting appropriate scales, baselines, formatting, and color schemes, eliminating the need for explicit user configurations.

The system automatically manages sizing, spacing, labels, and layout, ensuring charts remain readable even as data cardinality and density fluctuate. This capability is crucial for visualization intermediate language tools aiming for broad adoption.

A core strength of Flint is its ability to compile a single specification to multiple backends, including Vega-Lite, Apache ECharts, or Chart.js, without requiring a complete rewrite. This flexibility allows developers to choose the best rendering engine for their needs.

Flint's design is particularly suited for AI agent chart generation. Semantic types are often easier for models to infer than intricate visualization parameters, reducing the likelihood of errors.

Internal studies by Microsoft Research demonstrated Flint's efficacy, showing higher LLM-judge scores for chart generation compared to direct Vega-Lite outputs across various GPT models. For instance, GPT-5.1 achieved a score of 16.27 with Flint versus 15.91 directly.

Flint is now integrated into Microsoft Research's Data Formulator project for AI-assisted data analysis. The open-source project includes the flint-chart library and the flint-chart-mcp server, enabling agents to create, validate, and render charts directly within chat or coding environments.