The first version of the system my team built was, functionally, a predefined component library.
It came out of a design-to-code effort, so the rules were never written down as rules. They were baked into code that already existed and already worked, and if you wanted to know what was allowed you looked at what had already been built. That’s an extremely efficient way to answer the question and it’s also the whole problem.
Developers loved it, and of course they did, because the expensive part was finished and what was left was translation, taking something already sound and fitting it to whatever a given platform needed. For them it paid off on contact.
Creative never really showed up, and the reason they gave was that the rules were too limiting, which I treated as a taste problem for longer than I’d like, assuming people would come around once they saw the speed. It wasn’t a taste problem. The library genuinely could not hold what they wanted to make, because it was thin on interactive components and had no room for the kind of exploration that makes a campaign feel like anything, and every idea had to arrive already shaped like something we’d shipped before. We had asked the creative work to start from the inventory.
Underneath that sat a question we didn’t answer for a long time, which is whether the exploration leads or follows. In 1.0 it followed, so you looked at what existed and you designed into it. The alternative is a partnership where the creative vision gets to shape what the components should be, and the system’s job is to catch that vision and make it repeatable rather than to have pre-decided it.
We also drew those lines by instinct, with somebody experienced looking at the work and making a call about what was structural and what was decoration, and sometimes that call is right, but it’s never checkable.
Governance is a set of doors sized to the people walking through them
The instinct with the word governance is to picture a gate. One checkpoint, somebody with a clipboard, a queue. That’s why designers hear it and reach for their coats.
The version that actually works looks less like a gate and more like a shop where the tools are already laid out for whoever walked in.
A motion designer should be able to change a duration or an easing curve without knowing how a token resolves, or what that value becomes in an email client, or which four components inherit it. Those are real concerns and they belong to somebody. They don’t belong to the motion designer, and making them theirs isn’t rigor, it’s the system handing its own homework to somebody who came here to do a different job.
A copywriter should be able to find out that the headline wraps to two lines at 320 without learning the type scale. A campaign designer should be able to reach for a different composition without wondering whether they’ve just broken the disabled state of a button three screens away.
The goal I’ve landed on is that creative teams shouldn’t need deep system knowledge at all. They need to know how to describe what they want to the people who hold that knowledge, and they need those people to be close enough to answer. Somebody still has to make the primitive-level decisions and keep the token mapping honest. That work is real and it is somebody’s whole job. It should never quietly become the creative team’s job under the banner of empowerment.
The system’s job is to hand each discipline the handles that belong to their craft, hold everything downstream of those handles steady, and never make anyone learn a neighboring discipline in order to do their own. Good governance feels like a good partner rather than a good bouncer. It gives you what you needed before you asked, it maintains the hundred relationships you don’t want to think about, and it stays out of the way on the decisions that are genuinely yours.
The lock is what makes the loose part possible
The underdrawing is the sketch beneath the painting. It fixes the composition, the proportions, where the light falls, all the structural decisions that would be miserable to relitigate once there’s paint on the panel. And precisely because those are settled, the paint on top gets to be loose. Sargent’s brushwork looks reckless from six inches away and it works because none of the recklessness is load bearing.
That’s the argument for locking, and it runs opposite to the way people usually hear it. You fix the composition so the brushwork can be free. Fix nothing, and the looseness ends up carrying structure it was never built to carry.
This is the creative decision, and it needs a particular person
Deciding what to fix and what to free is the most creative decision in the entire system, and it gets treated like maintenance.
To make the call well you have to hold two things at once. You have to know what the brand is actually for, which parts carry meaning and which parts are just habit, where the recognizable thing really lives. And you have to know what the machinery does, how a value propagates, what it costs to open a handle, which openings are safe and which ones quietly become twenty-three grays in eighteen months.
A brand lead has the first half. An engineer has the second. Neither one can make the call alone, and when you split it, you get a system that’s technically immaculate and creatively dead, or one that’s beautiful and unmaintainable. This is the role from the last issue, the one without a good name, and this decision is the clearest argument I have for why it needs to be a role and not a standing meeting.
How I actually decide
There’s no universal ratio, but there are four questions I keep coming back to.
- Does it carry meaning or does it carry expression? Lock the first, open the second. The action color means something, and being wrong about it costs trust. A composition should mean something different every time, and being the same about it costs interest.
- Can the person holding this handle explain what it does? If they’d have to learn another discipline to answer, it isn’t open, it’s a trap with a friendly label on it. Close it and give them something they can reason about instead.
- Has this been an exception twice? Then it’s a decision you haven’t made yet. A standing queue of one-off approvals is you making the same design call one ticket at a time and calling it governance.
- What breaks if this is different everywhere? If the honest answer is “nothing, it just looks different,” you’ve probably locked it for tidiness. Tidiness is not a business case.
Run those four at the system we built back then and it fails most of them. We locked composition because leaving it open was expensive, not because it carried meaning. We handed creative a set of handles they couldn’t reason about without learning our half of the job. And when the exceptions came in, we processed them.
What’s changed since isn’t that I got better at knowing where the line goes. The ground moved. Code that used to be the costly part is fast enough now that most of what we locked in 1.0 was only ever locked because rebuilding it was expensive, and that reason has quietly expired. So I’d open more of it today, and I’d open it earlier, and I’d stop treating the existing inventory as the starting point for anything.
Not all of it, though. The system was never meant to capture everything, and I’ve stopped waiting for the version that does. There will always be work that wants to happen outside the structure, and the structure’s real job is to make that a decision somebody makes on purpose rather than an accident nobody notices until the fourth gray.
What makes the open parts survivable is people. Documentation doesn’t do it and neither does a better diagram of the token architecture. It takes somebody sitting close enough to the creative work to catch the thing that wants to exist and fold it back in.
The other reason I’d open more of it now is that I’ve stopped being nervous about what’s arriving, which is a strange thing to say in a year like this one. Everything new shows up above this layer. The generators make the component or the page or the whole campaign, and every one of them has to stand on some existing set of decisions about what the blue means and when it’s allowed to appear, which is the thing we’ve been down here building the entire time. New tools don’t take that from you. They consume it, and the better they get, the more they need it to already be right.
That’s the next thing I want to write about, because the part that keeps this layer yours isn’t that it sits deep.
It’s that somebody still has to decide what any of it means.