Design Systems as Agent Context

We came across Magnus K. Hillestad's thinking on the five layers of agentic context. It struck us as a useful way to think about where organizations are headed.

Because "give the agent context" is starting to sound a lot like "give the agent data." Those aren't the same thing. An agent can have access to your data and still not understand:

• What your company stands for.
• What it's allowed to do.
• How things are supposed to work.
• Which rules matter.
• What decisions have already been made.

That's where canon gets interesting. The approved ways of working within an organization include products, policies, brand, design, content, and rules. These become much more important when agents start acting on the organization's behalf. They're no longer just things people reference. They become context the machine has to understand.

And that has some pretty interesting implications for design systems. Because a design system is more than documenting how a button works. It captures decisions about how an organization builds and communicates things like what's approved, what's reusable, what patterns belong, and what the experience is supposed to be.

That's a very different role for a design system when AI starts building. We're starting to see the pieces of this emerge through things like MCP, agent-aware design tools, component documentation, and machine-readable design tokens.

The bigger shift, at least in our little corner of this world, is that the design system is moving from documentation to infrastructure. Something people, software, and AI can build from.