Don't preserve existing UI surfaces out of comfort or sunk cost. When the user archetype shifts, rebuild the interface to reflect changing tides.

In creative writing, there's a piece of advice every writer eventually hears: kill your darlings. It means cutting the sentences you love most when they're no longer serving the piece. Sometimes the thing you're most proud of is what you need to cut.

Interface design has the same trap. We ship something and refine it. Users learn it and the team comes to know it intimately. The UI accumulates gravity not because it's right, but because it's what we're used to.

Consider a product that starts out built for beginners: assistive, step-by-step, full of guardrails. The user is still driving the work; AI is copilot.

Eventually, AI gets good enough that beginners stop being the user. The people now reaching for the tool are experts, with different mental models, different trust requirements, and a completely different sense of what good output looks like.

You can't retrofit a beginner's interface onto an expert. The metaphors, defaults, and confidence level the UI projects don't fit anymore. It's time to start over.

In AI-native products especially, interfaces are not durable artifacts. They are current best guesses at how to surface capability. As capability compounds and user archetypes shift, our interface hypotheses expire.

An interface that doesn't fit the new user creates a ceiling. The product stops growing into what it could be.

I think about this a lot at Paradigm. Our current interface is built around a grid: analysts composing agent workflows cell by cell, with visibility into each step. That's exactly right for where we are: making agents legible, building trust, giving users control as they learn to delegate.

What does the interface look like when nobody is touching the grid anymore?

I don't know yet. But I have a feeling we'll have to kill something we love to find out.