Knowing what you need and being able to build it are two different skills. The distance between them is where a lot of ideas end up dying on the vine. However, on a recent project, I managed to bridge that distance.
In the middle of a rebranding project where the destination was an elegant, elevated, and clear visual expression, I proposed a light-touch gradient to use as an accent across different surfaces. I created the platonic ideal of the gradient and envisioned functionality to further enhance its presence as part of the brand.
The Gap
Creating the example exposed a key question: how do I make this gradient consistent and extendable? The rebrand work would establish the framework for future expressions, but it was unrealistic to hand-design every iteration. What would this look like if someone other than me created these? Adaptability and consistency pull in opposite directions: one needs to flex across new surfaces while the other needs to hold the line.
In another version of this story, I pop into the project Slack channel with context and a brief for an engineer. Or I keep exploring with that balance between consistency and versatility. I knew the rough shape of what would solve the problem: something that could generate brand-true gradients with specific adjustment levers for different instances. I could describe it, but I could not, on my own, build it.
What I Did
Working with Claude Code, focusing on the functionality I wanted for myself as well as a future non-designer, I generated a mesh gradient tool. The specs themselves matter less than the fact that every one of them is a brand decision wearing technical clothes. The guardrails I wanted were the brand strategy expressed in a form people could actually use.
Generating the mesh gradient tool is beside the point compared to what it did: collapsing the distance between the idea and the thing to almost nothing.
What I Didn't Do
I’ve been thinking a lot about this experience, because the simple breakdown is that I just used and LLM to make a gadget. What I believe really happened was that I experienced the next iteration of what it means to design brands, interfaces, and products.
The LLM did not suggest the tool. It did not nudge me toward building one, or surface the gradient as a problem worth solving, or design the system of constraints that made it brand-relevant. All of that came from the previous strategic work, from looking at what we'd made and recognizing the specific shape of what was missing.
What Changed
The old version of this hand-off is a guideline: a document that describes how the gradient should behave. Brand governance, in that model, is something an agency polices from a distance and a client slowly drifts away from.
What we gave instead was a live tool they now own and operate. The rules aren't written down somewhere to be remembered and eventually ignored; they're literally encoded. Governance stops existing as a document to enforce and becomes a built-in capability.
What feels new isn't that I can make more, faster. It's that the form of what I make shifts from instructions to instruments in a way that deepens a client's ownership of their own brand.
What Happens Next
For a long time, the question I ran into was can I build this? It shaped what I let myself propose and limited what I might even attempt on a project. That question has changed both in wording and timing. Now, it’s do I know what's worth building? And it occurs much earlier in strategic thinking.
So much more can be built at a previously unimaginable pace. We owe it to ourselves, our clients, and users to take those extra steps to consider what we are building, and why. Increased capability and speed are nothing without the right direction.
The less I have to worry about whether something can be made, the more it falls to me to know what's worth making. To look at a brand and see the one small thing that would make it stronger, and to have the restraint to build only that. The tools collapse the distance between an idea and the thing. They do nothing to help me find the idea.
