The Guardrail Problem

AI makes you fast. Dangerously fast. A layout that used to take an afternoon collapses into minutes. A color system, a component library, a full page. Generated, iterated, discarded, regenerated. The speed feels like freedom until you look up and realize you have no idea where you are.

I've been there.

Figma files ballooned. Frames stacked on frames. What started as a clean grid became sediment: every AI assist left a layer. Nothing was wrong, exactly, but none of it felt quite mine, either.

Code slows you down the right way. It asks you to mean it.

Working in code feels different.

The syntax resists you. You have to name things, structure things, decide what a rule is before you can write it. That friction forces you to be precise. You can't accidentally leave twelve versions of a button in a stylesheet. The compiler keeps score, and so do you.

What helps is building constraints that let you run.

Tokens enforce a single source of truth. Agents watch the work and flag the drift — the redundant style, the off-grid frame, the third shade of gray that crept in. Let AI do what it's fast at, and let code enforce what it's precise at.

We talk about guardrails as something to keep the AI in check. What I've found is that I'm usually the one who needs them more. I get excited. I keep generating when I should stop and read what I have. The AI doesn't need to take a breath. I do.

logo.svg

Jacques Ramphal

I lead design work where systems, code, and AI meet—building the practices and platforms that let cross-functional teams deliver meaningful products efficiently and sustainably.