Speed stopped being a back-office engineering concern a while ago. In 2026, Core Web Vitals sit squarely on the line between a good experience and lost revenue — and INP is the one that still trips teams up.
Speed stopped being a back-office engineering metric a while ago. In 2026, Core Web Vitals sit squarely on the line between a good experience and lost revenue.
Three vitals still anchor the conversation: Largest Contentful Paint for loading, Cumulative Layout Shift for visual stability, and Interaction to Next Paint for responsiveness. INP — which replaced First Input Delay in 2024 — is the one that still trips teams up, because it measures the full latency of every interaction, not just the first tap.
LCP and CLS are largely about how a page arrives. INP is about how it feels once you’re using it. A heavy main thread, oversized JavaScript bundles, and expensive event handlers all show up here, in the lag between a click and a visible response. It rewards exactly the server-first, ship-less-JavaScript discipline the rest of the industry is moving toward.
The fix is rarely a single trick. It’s breaking up long tasks, deferring non-critical work, and being honest about how much code runs on interaction. Measured on a mid-range phone, the difference between a sluggish and a snappy interface is usually a few hundred kilobytes of JavaScript nobody needed.
Vitals correlate with the numbers leadership cares about. Faster, more stable, more responsive pages reduce friction in the funnel — and the compound effect of better rankings plus higher conversion can move revenue meaningfully. That’s why we treat performance as a product requirement, not a cleanup task.
Practically, we prioritise by impact and coverage: improve the templates that serve the most high-value traffic — product, category, and service pages — before chasing long-tail URLs.
Don’t guess — measure. Then fix the template that touches the most revenue, not the page that’s easiest to reach.
Performance regresses the moment you stop watching it. We wire vitals into monitoring so a heavy third-party script or an unoptimised hero image shows up as a number, not a complaint. A site that launches fast and rots in six months was never really built for speed — it was just new.
The goal isn’t a perfect lab score. It’s a page that respects the person’s time and device, every single visit.
Theo writes about the engineering side of the studio — performance, architecture, and building things that last.
For years we shipped the whole application to the browser and apologized for the spinner. In 2026 the default is server-first — and the client is finally getting lighter.
Read article
AI has gone from autocomplete to collaborator. Here’s how agentic tools are reshaping the way a studio researches, explores, and ships — and why a designer’s judgment matters more than ever.
Read article
A static interface assumes every user wants the same screen. Generative UI starts from the opposite premise — and quietly moves the designer’s job from drawing screens to defining the rules that build them.
Read article