Beyond static charts: Helping AI create visualizations you can explore

AI agents can generate static data visualizations to analyze patterns, but exploring data interactively—such as zooming, comparing charts, or linking selections to text—remains challenging. Microsoft’s new Flint-interactive language addresses this gap by providing a high-level way to specify interactions, making it easier to create dynamic charts without complex coding. The system leverages Flint’s compiler to translate user interactions into semantic events, simplifying the process of building interactive dashboards and data stories.

Flint-interactive integrates with existing tools like Flint-MCP and MCP Apps, allowing users to instruct an agent to generate both a chart specification and an interaction specification directly within chat or canvas environments. This enables the agent to render interactive charts dynamically based on user requests, reducing the manual effort required to build such visualizations. The approach separates visual logic, handled by the compiler, from interaction logic, specified at a semantic level, making maintenance and updates more straightforward.

For developers, Flint Canvas provides a state machine to track interactions and manage visual rendering, while a programming API allows for custom interactions beyond preset options. Event handlers can process semantic events to update charts, tables, or application state, enabling connected views in dashboards and data stories. The system’s compositional design ensures that interactions can be easily composed with other elements like tables or text.

Microsoft positions Flint-interactive as a tool for chart users, app builders, and interaction designers. Users can set up Flint in chat and request interactive charts, while developers can build dashboards with the programming API. Designers can create bespoke interactions and propose new presets on GitHub, expanding the system’s capabilities and fostering community contributions to its interaction vocabulary.