If you learned React before 2023, there's a decent chance some of what you know is now actively getting in your way. The introduction of Server Components, the React Compiler, and deep framework integration with Next.js has changed not just the API surface but the *mental model* you need to use.
The biggest mental model shift: where does this code run?
The defining question of modern React isn't "should this be state or props?" It's "does this code run on the server, the client, or both?" Server Components run once on the server, have no event handlers, can query databases directly, and produce pure HTML. Client Components are what you've always known: they run in the browser and can use hooks and event listeners.
The default in Next.js 14 App Router is now Server Components. You opt into Client Components with `"use client"`. This is backwards from what many React developers expect, and it's the source of a lot of confusion in migration.
The React Compiler changes how you think about memoization
For years, React developers reached for `useMemo`, `useCallback`, and `React.memo` to prevent unnecessary re-renders. The React Compiler (now stable) automatically applies the equivalent of these optimizations during the build step. In most cases, manual memoization is now either unnecessary or actively counterproductive.
This doesn't mean you never optimize; it means you optimize *after profiling*, not pre-emptively. Write the simple, readable version first. Trust the compiler. Measure. Only then reach for manual optimization if the profiler shows you a real bottleneck.
What actually matters in 2026
The skills that transfer regardless of React's evolution: understanding the browser event loop, knowing when and why re-renders happen, thinking in terms of data flow rather than component structure, and having a clear mental model of the network boundary. Frameworks change. These fundamentals don't.
The React ecosystem in 2026 rewards engineers who understand *why* things work, not just *how*. If you've been copy-pasting patterns without understanding them, now is the time to go deeper. The component model is simpler than it's ever been, but the mental model required to use it well is more nuanced.