Microsoft Research has introduced a chart-interaction language called Flint-interactive. AI agents can describe how charts respond to clicks, selections and other actions. It connects visual elements with the underlying data so an interaction can update charts or related text. The primary announcement describes a compilation system and shared interaction state, with examples aimed at dashboards, data stories and embedded visualizations.
Microsoft Research, Microsoft’s research organization, has introduced Flint-interactive for AI agents that generate data visualizations. The single primary announcement describes a language built on Flint, its visualization system. An agent specifies what should happen when a user clicks, selects data or changes a view, while a compiler handles the connection to the chart’s rendered elements.[1]
The compiler translates mouse and keyboard actions into semantic events: a description of which data the user has selected. Update operators then highlight elements, add annotations or change the viewport. In Microsoft’s examples, selecting data highlights related elements across charts or connects the chosen points with explanatory text. Flint Canvas, the component managing selection and display state, lets different parts of an application share that state.[1]
Flint MCP, a connection tool for AI applications, links the visualization system with chat software. MCP Apps can deliver an interaction specification with a chart. Larger applications can connect charts with tables, widgets and text, while custom handlers extend the available interactions. Microsoft describes dashboard and data-story uses; the announcement describes the system’s design rather than establishing independently measured gains across every agent model or visualization task.[1]
