e observers, no z-index fights. The browser handles collision detection and fallback placement.
3. View Transitions — page animations, zero JS
Cross-document View Transitions: Chrome 126+, Safari 18.2+, and an Interop 2026 target (Firefox is implementing).
@view-transition {
navigation: auto;
}
::view-transition-old(main) {
animation: fade-out 0.2s ease;
}
::view-transition-new(main) {
animation: fade-in 0.3s ease;
}
Enter fullscreen mode Exit fullscreen mode
Full page-to-page navigation animations with plain <a> tags. No client-side router required.
4. :has() — the selector most people still aren't using
Universally supported since December 2023. Lets you style a parent based on its children — something CSS literally couldn't do before.
.form-group:has(input:invalid) {
border-color: red;
}
.container:has(.item) .empty-state {
display: none;
}
main:has(+ aside) {
grid-template-columns: 1fr 300px;
}
Enter fullscreen mode Exit fullscreen mode
Teams report deleting hundreds of lines of class-toggling JS after adopting this properly.
Baseline 2025:
.parallax-element {
animation: parallax linear;
animation-timeline: scroll();
}
@keyframes parallax {
from { transform: translateY(100px); }
to { transform: translateY(-100px); }
}
Enter fullscreen mode Exit fullscreen mode
GPU-composited, tied directly to scroll position. No requestAnimationFrame, no jank.
6. Next.js 16's "use cache" — the model most upgraders miss
"use cache"
async function ProductPrice({ id }: { id: string }) {
const price = await getPrice(id);
return <span>{price}</span>;
}
Enter fullscreen mode Exit fullscreen mode
Per-component cache control instead of per-page. A single route can have a statically cached header, a dynamic price, and a streaming reviews section — each with its own lifetime. Combine with Partial Prerendering (cacheComponents, no longer experimental).
7. React Server Components — the numbers, one year in
- Bundle size: 40-60% reduction in real codebases
- LCP: ~22% median improvement over Pages Router
- Turbopack (now default): 10x faster cold starts
The failure mode nobody warns you about: RSC isn't a performance toggle, it's an architectural boundary. Teams that redesign their component tree around the server/client split succeed. Teams that sprinkle "use server" on existing components fight the model the whole way.
8. Agentic coding — past autocomplete
The real shift isn't Copilot suggesting the next line. It's agents that read your codebase, plan an approach, write across multiple files, run tests, and commit — Claude Code, Cursor, Cline.
What works: Explore → Plan → Code → Commit, on a bounded scope ("add dark mode to settings").
What doesn't: "Build me a SaaS." Agents excel inside an existing codebase, not greenfield architecture.
9. WebAssembly at the edge
Rust, Go, and Python now run on Cloudflare Workers, Vercel Edge, and Deno Deploy — compute-heavy work (image processing, PDF generation, data transforms) without cold-start penalty.
Also worth knowing: Vercel Edge runs on Cloudflare's worker infrastructure under the hood, Netlify and Supabase edge run on Deno Deploy. The platforms are converging, so your edge code is more portable than it looks.
Bottom line
Every feature above is production-ready today, and most can be adopted incrementally — no rewrite required. The teams pulling ahead in 2026 aren't chasing the newest framework. They're deleting the code the browser now handles for them.
I write about this stuff at ProWebify — we build with Next.js, React Server Components, and edge-first architecture. Questions or disagree with something above? Drop a comment.
