Teads Engineering: how AI let designers write their own components

Matthieu Rebuffat, a senior product designer at Teads, describes a bottleneck that many design teams recognise: designers ship components faster than developers can code them. The backlog grows, frustration accumulates on both sides, and the design system falls progressively behind the product it is supposed to document.

His team’s fix was direct and tool-driven. By connecting Figma with an AI-powered IDE through an MCP connector, designers gained read-write access to the codebase. They could write component code themselves — without first completing a formal engineering education. What previously took several weeks to clear the developer queue could be turned around in a couple of days.

The shift changes the designer’s relationship to the repository in a fundamental way. Instead of producing specifications and waiting, Rebuffat’s team started handling simpler tickets that developers had always deprioritised due to bandwidth constraints. The AI handled mechanical code generation; the designer supplied the product and visual context that the AI lacked. The collaboration moved from a sequential handoff to something closer to a shared workspace.

Rebuffat also draws a careful conclusion about what this makes possible. When designers can close simple code tickets themselves, engineering time frees up for harder problems. Velocity is no longer the differentiator when everyone can ship faster — quality of decisions and depth of problem-solving become the advantages worth building. The article argues that companies should resist using newfound speed to simply produce more output, and instead use it to produce better work.

The piece is short, grounded in what one person observed at a specific company, and free of speculative extrapolation. That makes it more useful than most workflow posts in this space. Rebuffat acknowledges the limits of the approach: the method works for designers handling simpler tickets, not for complex architectural decisions that still require engineering expertise.

Useful for design system managers evaluating whether to give designers repository access, and for teams considering MCP tooling to reduce cross-functional handoff delays.