Frame 2049519452.jpg

The New Workflow

Yesterday, Figma introduced Claude Code integration, where code flows to Figma and back again in a continuous loop. This is important because the loop works both ways; your team creates in Claude Code using your structured brand files, sends it to Figma, your designer refines it on the canvas, then brings it back to code with Figma MCP.

Certain files (like brand-positioning.yaml and messaging-framework.md) serve as the single source of truth for both sides - the agent reads them when generating code, and the designer references them when refining in Figma. They both employ the same positioning, same voice, and same structure.

However, this only works if your brand has structured data, not just Figma files.

At Little Plains, we’ve been preparing for this over the past few months. We've been slowly shifting from just delivering brand guidelines to delivering brand infrastructure. I think it's cool and hopefully this is helpful.

Here’s how we're approaching this:

Brand Infrastructure

We've been working on producing two layers for each brand:

It's less about replacing one with the other, and more about giving teams what they need, however they're working. Your designer needs Figma files and your agent needs structured data, and both need the same positioning, voice, and structure.

First: The Human-Facing System

This is the craft: visual identity, typography, motion, website, narrative, and brand guidelines. PDFs, Figma files, design systems; this is what people see and feel. This is what we're 'used to' delivering, and receiving.

Screenshot 2026-02-17 at 10.42.19 PM.png

A bird's eye view of a digital brand guideline (for humans).

Second: The Agent-Readable System

This is the structure. Markdown, YAML, and JSON files that define positioning, voice, differentiation. It’s for your team to actually use when they're working with agents (which increasingly, we all are).

We call it an Agent-First Brand Kit, a structured knowledge system where every piece of brand guidance lives in a chunk (one of my favorite words ever), approximately 400 tokens each, with metadata that agents can parse and humans can read.