Treating Core Web Vitals as a budget
Performance is not a phase at the end — it is a number you spend against from the first commit.
By Axicial
If performance is something you 'get to later', youA"We'll optimize performance at the end" is how every slow site got slow. Performance debt behaves exactly like technical debt — it compounds quietly, one dependency and one hero image at a time, until the optimization pass at the end isn't a pass, it's a rewrite. The teams that ship fast sites don't optimize harder. They just never let the number move in the wrong direction in the first place.
Pick the number before you pick the framework
A performance budget only works if it's a real constraint, not a suggestion. We set three numbers before any design work starts: a target LCP (largest contentful paint, usually under 2.0s on a throttled mid-tier connection), a target INP (interaction to next paint, under 200ms), and a JS bundle ceiling for the initial route, usually somewhere between 100-150KB gzipped depending on the product. Everything downstream — framework choice, image strategy, how much gets client-rendered versus server-rendered — gets negotiated against those numbers, not the other way around.
This changes decisions early in a way that's hard to retrofit. A carousel that needs a 40KB library gets weighed against the bundle ceiling before anyone's attached to it emotionally. That conversation is cheap in week one and expensive in week twelve.
Enforce it in CI, not in a Lighthouse tab someone forgets to open
A budget that lives in a Notion doc gets violated by the first urgent feature. A budget enforced in CI doesn't. We wire bundle-size checks and Lighthouse CI runs directly into the pipeline with hard fail thresholds — a PR that pushes the initial bundle past the ceiling doesn't merge until someone either trims it or explicitly overrides it with a reason attached. The friction is the point. It forces the "is this feature worth the weight" conversation to happen at PR time instead of at the next quarterly performance retro nobody schedules.
The unglamorous 80%: images and fonts
Nobody's performance problem is really the JavaScript framework. It's almost always images and fonts, because those are the assets that get added by whoever's least likely to think about bytes — marketing, content, a client uploading their own hero shot. We standardize on an image pipeline that auto-generates responsive srcsets in modern formats (AVIF with WebP fallback) at build or upload time, so the correctness of the output doesn't depend on someone remembering to compress a file before uploading it. For fonts, that means subsetting to the actual character set in use, self-hosting instead of a third-party font CDN round-trip, and font-display: swap as a default, not an afterthought someone adds after noticing flash-of-invisible-text in production.
CLS is a layout discipline problem, not a performance problem
Cumulative layout shift almost never comes from "slow" content — it comes from content whose dimensions aren't reserved before it loads. Every image needs explicit width and height (or an aspect-ratio box) before it ever ships, every ad slot and embed gets a reserved container, and web fonts get a fallback stack sized to match the metrics of the real font so text doesn't reflow when it swaps in. This is a code review checklist item, not a Lighthouse fire drill — catching it at review time costs a comment; catching it after launch costs a debugging session against real user data trying to figure out which component is the culprit.
What the budget actually buys you
The business case for this isn't abstract. Field data — from real users, not lab conditions — is what search ranking and conversion both respond to, and lab scores that look great in a staging environment routinely diverge from what real users on real networks experience. A budget enforced from commit one is cheaper than a performance sprint bolted onto the end of a project, and it produces a number the team can actually defend, because it was never allowed to drift in the first place.
Performance work done at the end is always a negotiation with a deadline. Performance work done as a budget is just... engineering. never do. We treat Core Web Vitals like a budget set on day one, tracked in CI, and defended on every pull request. Here is the setup we reuse on every infrastructure engagement.
Want this kind of work on your project?
Tell us what you’re building and we’ll map the fastest path from idea to launch.
Start a project →