Agents are only as good as their context

The Story

Figma’s 26 August workflow lab explores movement between design and code with agents, focusing on keeping the canvas and implementation in sync. It sits alongside the company’s wider emphasis on MCP, design systems, and agent-assisted product work. ([figma.com](https://www.figma.com/blog/ai/?utm_source=openai))

The useful principle is that agents do not magically understand design intent. They need the vocabulary of the system: components, states, naming, constraints, accessibility decisions, and the reasons behind them.

For a small studio, this suggests a practical investment: document the “why” behind your reusable patterns. A compact design-system file with examples and rules may become more valuable than a larger library of disconnected components.