Working Surfacelink · 4 Nov 2026
06
Codrops

Building Digital Products with Design Systems and Agents: A Guide for Terrified Designers

Jakub Otcenas · 1 November 2026

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

Read the original
Takeaway

Designer Jakub Otcenas of the studio Diorama builds its design system with AI agents, turning his comments from each session into a written rule book the agents follow.

Summary

Jakub Otcenas is a designer at Diorama, a design studio building bydiorama/ui, an open-source design system meant for AI agents from the start, for internal tools and client prototypes. He works in the design tool Paper with Claude Code, Anthropic's AI coding agent. He found that agents left to design alone produce work that feels empty, so he directs them as he would a junior designer.

Key points
  • He leaves feedback as comments on the design canvas, with screenshots and mood boards, explaining why each choice matters. He then asks the agent to apply the comments in a new frame.
  • At the end of each session the agent turns the thread into rules and skills. A file called CONVENTIONS.md holds the rules, including bans on typical agent habits such as thick left borders on cards.
  • Agents do the structured work: token systems, accessibility checks and documentation of each component's states. He redraws by hand where the generated versions miss, as with a banner none of them got right.
  • When an edit fails in code, he has the agent bring the design back into Paper, fixes it there and sends it back to code.
Implication

A design team can build its agents' rules from recorded review comments rather than writing them all up front. This is one designer's account of a small studio's practice, with no measured results.

Plays
Design

At the end of a working session with an AI agent, have it turn the designer's review comments into written rules or skills that later sessions follow.

Design

When a change made with an AI agent goes wrong in code, fix the design in the design tool first, then have the agent carry it back into code.

Derived by Working Surface from the article; more in the Playbook. Source line: Outsourcing all your thinking to agents isn’t a valid strategy.

Source issue

4 November 2026: teams building with AI agents are turning each failure or correction into a permanent check, rule or shared record, rather than fixing one result at a time.