Stuart Frisby nails a problem I’ve been observing for a while.

This is where design as infrastructure stops being a slogan and becomes a requirement. If the tool can’t communicate a standard, the standard has to be reachable from any tool. That means the context a designer needs, the system they build within and the operating model that says how decisions get made all have to exist at a level of abstraction that a sketching tool, a prompting tool and a coding agent can each read. Tokens and components are the easy part. The harder work is the layer above: why we make the choices we make, what the product is for, what it should never feel like. Right now that lives in people’s heads and in documents nobody opens.

I’ve made the point before that AI is the primary future customer of the design system. I want to extend that. The customer is any tool a designer picks up, and there are going to be hundreds of them. A system that can only be consumed through one interface is a system with one customer, and that customer won’t last long, either.

I’ve long had the view that a design system (as we’ve known them) is one consumable form of a design language. A design language, and the context it provides for all sorts of tools, lies in a context layer. Always has, really.

Take a corporate design system. Covering branding, physical product design, wayfinding, digital products. Multi discipline. One brand. What this documentation typically covers is fundamentals and examples. These fundamentals are represented at level of abstraction. The examples point the way, but are often out of date before they are published. The truth is, the work in the wild is the truest representation of any system in practice. A context layer would hold this information. Updated in realtime.

This starts to solve the one problem any design system, brand system, etc has. In their attempt to be a canonical source of truth, they capture a moment in time. From then on, the process of evolution lies in conversation, in the work, in fragments of code of new features.

A context layer would be super helpful for teams. It captures a snapshot of the actual work happening, the decisions that are leading to it, and a constantly updating reflection of your design language zeitgeist. From there emergent design principals would form, new patterns emerge, and outliers identified.

Does this change the role of a designer in this AI world? No, I don’t think so. I think we have a choice. This approach could certainly provide everything AI would need to make a feature based on what it sees. Where this could get interesting if there was inbuilt recursive loops to self improve. But I reckon those are some thoughts for another time before I get into a debate about taste or judgement and what it means to be human.