Working Surfacelink · 3 Nov 2026
03
A single carved woodblock rests on a hand press while a long fan of near-identical printed sheets spills from it across the floor, the whole scene drawn with wide empty margins on every side.
UX Collective

State of a system

Pekka Oskari Puhakka · 2 November 2026

How-to · design systems · decisions · Agent context files

Read the original
Takeaway

Design consultant Pekka Oskari Puhakka, recalling a Figma library of 1,786 variants, argues that rules governed as an AI skill could keep a design system current.

Summary

Pekka Oskari Puhakka, a design consultant, wrote in UX Collective about merging a client's separate design kits into one company-wide library in Figma, the design software. A council of designers from each business unit made the decisions in the work, which began in 2022. The delivered kit held 1,786 component variants and was demanding to govern.

Key points
  • The working file reached about 4,000 variants, and the button alone had 260, of which only 15 needed real design decisions.
  • Figma features released during or after the project, such as variables for theming and typography, could have cut the variants to as little as a fifth.
  • A council with members from each business unit needed time, a shared culture and a willingness to set aside favoured designs.
  • Puhakka argues that governing an AI skill could be a better basis for an evolving design system, and says a new project would be needed to prove it.
Implication

That design system teams can separate the few choices needing a designer from variants a rule can produce is an inference and is not the author's claim. The AI proposal is the author's untested idea, written in hindsight about a project that ended about three years ago.

Plays
Design

Before adding variants to a component library, separate the few that need a design decision from those that only repeat sizes, and handle the repeats with variables.

Derived by Working Surface from the article; more in the Playbook. Source line: Arithmetic can express a decision. It cannot make one.

Source issue

3 November 2026: teams working with AI agents are moving human review from each output to the plan, the shared instructions and the rules the agent follows beforehand, and naming in advance who answers for the result.