I've been thinking about design systems for a long time. At Embark, we build them for brands that need their identity to scale — across platforms, teams, and time zones. The biggest challenge has never been building the system. It's been getting people (and now, tools) to actually use it correctly.
That's why Figma's latest announcement stopped me in my tracks.
What Figma Actually Shipped
Figma just opened its canvas to AI agents. Not in a vague, "AI can help you design" kind of way — in a direct, write-to-your-file, use-your-actual-components kind of way.
The key piece is a new tool called use_figma, available through Figma's MCP (Model Context Protocol) server. AI tools — Claude Code, Cursor, Copilot, and others — can now create and edit designs in Figma using your real components, tokens, and variables. Not converting HTML into generic layers. Actually reaching into your design system and building with what's there.
The first time I read that, I had to re-read it.
The Feature That Changes Everything: Skills
But here's the part that really got me — Skills.
Skills are markdown files that teach AI agents how your team works. Naming conventions. Which component to use in which context. How to handle edge cases. Rules your senior designer holds in their head after three years on the project.
Figma launched nine community Skills on day one, including one for generating designs with your existing components and another for detecting drift between code and design tokens.
This is the thing I've been waiting for. Design system documentation has always been written for humans. Skills make it actionable for machines.
Your design system stops being a Notion doc someone has to remember to check. It becomes a set of instructions an AI agent actually follows.
What This Means for the Teams We Build With
At Embark, every brand and web project we deliver includes a design system — components, tokens, documented patterns. We build these things because they create leverage. One decision, made correctly, ripples across every future surface.
With Figma's AI agent support, that leverage multiplies.
Imagine briefing an AI agent to design a new landing page section. In the past, the output would be generic at best — it had no idea about your brand's spacing rhythms, your button hierarchy, your card treatment. Now, with a well-built design system and the right Skills, that agent designs in your system. It pulls your actual Button component. It respects your color tokens. It follows the layout logic your team established.
The brand surface stays intact. And it happens fast.
The Workflow That Excites Us Most
The loop that stands out to me:
- An agent designs a new component or section using your system
- It screenshots its output and iterates — because it's working with real components, not pixels, corrections stay structurally sound
- A developer picks it up and ships it, knowing the tokens and components map directly to code
That last step is critical. Because Figma's MCP integrates with the same tools developers already use — Cursor, Claude Code, Copilot — the handoff between design and code becomes less about translation and more about execution.
We've spent years trying to close that gap manually. Now the tools are starting to close it for us.
An Honest Caveat
This only works if your design system is actually built well.
An AI agent working with a messy component library is going to produce messy output — faster. The quality of what these agents can do is directly proportional to the quality of the system they're working within.
This is something we think about a lot at Embark. A design system isn't just an asset library. It's a decision framework. The better the decisions baked in, the better everything downstream — including AI-generated work — will be.
If you haven't invested in a real design system yet, this is the moment to do it. Not because it looks good in a pitch deck, but because it's about to become the foundation your AI tooling runs on.
Where We're Headed With This
We're actively exploring how to integrate Figma's agent capabilities into how we work and what we deliver for clients. A few things we're thinking about:
- Building Skills into every design system we deliver — documented conventions that aren't just for humans, but formatted for AI consumption
- Using agents for rapid iteration — briefing, generating, reviewing, refining within the system instead of starting from scratch every time
- Token sync as a first-class deliverable — making sure what's in Figma and what's in code stay in sync automatically, not through quarterly audits
This is early days. But the direction is clear. Design systems that are machine-readable will compound in value. Ones that aren't will become bottlenecks.
For the brands we work with, we want to make sure they're on the right side of that line.
