The system no one could see.

Delta had been an AKQA design partner for close to a decade. That relationship had produced something real — a shared design sensibility, a vocabulary nobody had written down, a way of working where the standards lived in people rather than documents. When it worked, it worked because the right people were in the room together. Then the organization grew and the rooms multiplied. Marketing started running its own visual production. Developer teams built employee-facing tools we’d never been close to. Each group had good intentions. None of them had access to the decisions behind what they were inheriting.

One plate = one affordance a team was working from. All seven are the same size and float at the same height, because all seven inherited the same thing. Under the first, four projector drops fall to the datum and the footprint is ruled long where it was set out — the decisions that located it, still there in one place. Under the other six there is nothing. Two groups are named because the record names them; seven is a legible field, not a count of anything the record gives.

The cohesion frayed. What had been held together by trust and shared judgment needed to become something that could outlast any one team, any one relationship. My working assumption was that the path forward was clear: document the decisions, codify the standards, give every part of Delta the logic behind every affordance so that teams we’d never met could still produce design that belonged to the same family.

The harder problem wasn’t design. It was literacy.

Even the teams we’d worked alongside for years didn’t fully understand what went into a considered design decision. The affordances they relied on — the ones that made their products feel coherent — were invisible precisely because the work had been done well. But that invisibility had a cost: there was no shared language for it, no framework for understanding why a component existed, what it was protecting against, what would quietly collapse if someone went around it.

One mark = one thing the system was holding in line. Nine is a legible field, not a count of anything the record gives. Nothing moves sideways between the two halves — only the ground they were standing on.

Before I could build a system, I needed to build the conditions in which a system would be valued. That meant months of workshops. Conversations that looked like education but were really about earning permission to codify something the organization didn’t yet know it needed. Nothing in any KPI captures that kind of work. But without it, the system would have been a document no one read.

My first approach was the direct argument. I made the case explicitly: here’s what a design system gives you, here’s what it costs to not have one, here’s the drift you’re already experiencing and what happens if it compounds. The argument was correct. It didn’t land.

Not because the organization was resistant, but because they had no frame for what they’d be agreeing to. A design system is an abstraction. You can’t feel the value of it until you feel the absence of it, and the absence was already normalized. I was asking them to invest in solving a problem they’d stopped noticing. I walked away from that pitch with a polite no. You can’t sell infrastructure outright. You have to create the moment where it becomes the only logical answer.

The four premises are the ones the case lists. Drawn flat on purpose: the relationship here is adjacency — a built thing, a barrier, nothing past it — and there is no depth in it to draw. The route that worked is a ghost because the case narrates it and never diagrams it.

I stopped trying to convince the whole organization and started looking for the moment where there was no alternative to working together — where every part of the business had the same deadline, the same problem, and the same need for a shared approach. That moment came when Delta collapsed four separate measures into a single loyalty currency.

The SkyMiles overhaul touched every part of the business that had ever touched loyalty. Every team had to change at once. Nobody could get there independently. I protected the proposal that put a shared design foundation at the center of that coordination effort, and the case made itself. Everyone needed a script for this moment. We gave them one that would still be useful when the project was over.

One block = one surface the change had to reach. The four measures are drawn unlabelled because the case names four and never says which four; the eight marks are a legible field, not a count — the case says every part of the business that had ever touched loyalty.

What started as a proof-of-concept for SkyMiles became the design infrastructure running across the Delta ecosystem.

Component adoption held across the channels that had been drifting worst. Teams that had been building independently now pull from the same foundation. The visual coherence that used to live only in rooms where the right people were present is now a property of the system itself — something that doesn’t leave when someone does.

What it didn’t solve: the literacy problem is ongoing, not closed. A system you can open is not the same as a system you understand. Teams still override components they don’t have context for. The workshops built enough shared understanding to get the system adopted; they didn’t permanently solve the gap between having a system and knowing why each decision inside it exists. That second problem — design education at organizational scale — is still open. I’d build toward it differently if I ran this again, probably by treating documentation as a first-class design surface rather than a delivery artifact.

The system outlasted the team that built it. That was the actual goal. Everything else is still in motion.