08/09/2026
Design systems used to be built for two audiences: designers and developers.
In 2026, there is a third one: AI coding agents.
While working on Pipeline, we saw this shift firsthand. Their engineering workflow makes extensive use of tools like Claude Code and Cursor, which means a traditional Figma handoff is no longer the whole story.
The system also needs to be understandable by machines.
So we started thinking about the Design System differently.
Components need clear hierarchy and predictable behavior.
Tokens need semantic names that translate cleanly from Figma variables into code.
Layouts need explicit rules, not just visual examples.
States, dependencies and conditional behavior need to be documented in structured, machine-readable formats.
The goal is not to let AI randomly recreate the UI.
It is the opposite.
Give AI enough structure that it can move fast without slowly destroying the product's visual and UX consistency.
For Pipeline, this meant combining the Figma Design System with structured Markdown documentation, semantic tokens and strict reusable layout patterns that Claude Code and Cursor could work with directly.
The result is a development environment where engineers can use AI aggressively while the product still behaves like one coherent system.
I think this changes the role of the Design System quite significantly.
It is no longer only a library of components.
It is becoming part of the product's machine-readable architecture.