The Hidden Instructions Your Design System Is Missing for Agentic Tools
In this segment, the presenter outlines the essential, often‑overlooked specifications a design system must provide to be usable by autonomous agents. He argues that token names should convey intent rather than raw values (e.g., moving beyond "blue‑100" to purpose‑driven identifiers) so agents can reliably select the correct visual token. He stresses the need for explicit component composition rules—clearly defined conditions, group‑level behaviors, and state triggers—that agents can read and act upon. The talk also calls for decision‑logic metadata such as slot definitions, character limits, min‑max counts, and nesting constraints to be encoded directly in the component specifications. Finally, he emphasizes that these instructions must exist both in the design representation (e.g., design‑tool artifacts) and the code representation, or in a shared medium, so agents have a single source of truth for interpreting and configuring components correctly.
