Flint-interactive: A language for creating interactive data visualizations

Microsoft Research has introduced Flint-interactive, a language built on top of its Flint visualization system that lets AI agents and developers produce charts people can explore by clicking, filtering, zooming, highlighting and linking views. The post, dated October 22, 2026, is credited to Chenglong Wang, Yuwei Xiao, Alper Sarikaya, Michel Galley, Chris Quirk and Jianfeng Gao.

The authors argue that even strong frontier models struggle with interactive charts: agents seldom get them right in one turn, repeated debugging burns tokens and time, and fixes can introduce broken interactions or poor layouts. Their answer is to split the work. Flint's compiler converts raw clicks, drags and key presses into semantic events that identify chart elements and data, and it handles hit testing and low-level rendering. Authors only map those events to update operators, such as setting a style, an annotation or the viewport.

Microsoft describes three routes for using it. Chat users can connect through Flint-MCP and MCP Apps, asking an agent to generate an interaction spec alongside the chart and carrying selections into follow-up questions. App builders can use Flint Interactive Canvas, where events can filter tables, update other charts or change application state. Designers can write custom handlers for bespoke behavior, such as a DimpVis example in which dragging a point along a time path shows how cities' temperature and precipitation change.

The significance, by Microsoft's account, is more reliable and cheaper interactive output, though the post offers no benchmark figures to quantify those gains.