AI summary by TwinMind
JavaScript Frameworks in 2025
Theo - t3․gg
54m 48s
2024 Framework Landscape Overview
- 2024 was evolutionary rather than revolutionary - no major new frameworks or paradigms
- Angular changed most significantly in 2024
- Most framework development driven by VC funding (Vercel/SvelteKit, Astro, Remix/Shopify, etc.)
- Only non-VC backed: Solid Start, Analog
Server-First Development Trends
- 5-year narrative shift toward server-first approaches
- Pandemic amplified need due to online shopping habits and low interest rates
- Battle between SPA-influenced isomorphic approaches vs MPA-influenced split execution models
- Key developments: Next.js app router, view transitions, server functions, optimistic updates, server islands, single flight mutations
- Out-of-order streaming notably absent from discussion despite importance
Technical Pattern Definitions
- Server functions: Client-callable functions that feel like local calls but execute on server
- Optimistic updates: Show updated state before server confirmation
- Server islands: HTML parts that load separately after initial page response
- Single flight mutations: Server responds with updated state in same request as mutation
Backend for Frontend (BFF) Pattern
- Every website benefits from having server layer optimized for frontend
- Even with existing backend APIs, thin JS server layer improves user experience
- Almost irresponsible to not use JavaScript on server for user-facing websites
- BFF enables better relationship between frontend and data/APIs
Complexity Reality Check
- Harsh reality: BFF + new routing patterns increase apparent complexity
- Average React dev doesn't understand underlying complexity of current SPA approach
- New patterns make previously hidden complexity visible
- Example: Dan Abramov didn't understand cold starts despite leading React education
Build Process and Tooling Complexity
- Webpack made build processes default, enabling exponential tooling complexity
- Going from codebase files to running app has become significantly more complex
- Edge worker deployment adds additional complexity layers
Performance Measurement Challenges
- SSR benchmarking failures across JS frameworks
- Performance wins often misattributed to new tech vs actual root causes
- Vercel's partial pre-rendering comparable to other edge solutions
- Image optimization often more impactful than framework optimizations
- Netlify Edge performs better than Vercel Edge due to Deno's globally distributed runtime
Traditional Server Approaches
- Return to non-SSR server approaches gaining traction
- Great choice for appropriate projects but uninteresting for complex needs
- Simplest tool correct when problem is simple, need options that scale
Compilation Revolution
- Compilation ubiquitous in JS development for handling obstacles
- JS Zero proposal: minimal JS core with compiler-added features
- Benefits: types, linting, tree shaking, code splitting, minification, isomorphism, macros, DSLs
- All advancements built on compilation foundations
React Compiler vs Svelte 5 Runes
- React Compiler: auto-optimizing compiler reducing unnecessary re-execution
- Adds useMemo/useCallback automatically, performs as well as best React engineers
- Svelte 5 Runes: syntactic sugar over fine-grain signal renderer
- Ironically opposite of original selling points for both frameworks
Framework Philosophy Evolution
- Change not pendulum but spiral staircase - back/forth while increasing quality
- React Compiler enables building Zoom-equivalent app as solo developer
- Tools enable larger teams (10-100 to thousands) working in same codebase
Complexity Ownership Models
- React Compiler: React team owns memoization complexity, developers own simple syntax
- Similar to GraphQL vs tRPC debate - who owns which complexity matters
- Compiler makes code easier for AI generation by removing memoization nuances
React as Final Framework Theory
- React Compiler solves ergonomic issues without changing core syntax
- Server components solve data management
- Constraining language vs expressing intent debate
- React syntax often didn't clearly communicate developer intent
- Compiler better encodes actual developer intent than manual memoization
AI Impact on Development
- AI impact on frameworks still minimal
- Devon created functional but not good applications (79 Lighthouse score, 130kb JS)
- V0 successful for prototypes
- React Scan reveals performance issues (e.g., Slack's constant re-renders)
- Development tooling rising to meet complexity needs
Single Page App Renaissance
- Swing back toward SPA models mid-2024
- SvelteKit, Solid Start, Remix porting non-server functionality
- Tanstack Start built on Solid's principles for React
- Next.js working on better traditional SPA support
- Local-first and sync engine tech increasing
Framework Sentiment Trends
- Growing dissatisfaction with JS frameworks
- Most tools showing declining sentiment (Preact, Angular pre-resurgence, React)
- Stable tools (Vue, Angular, OG React) shine during economic tightening
- Best tool isn't always the best tool - developer happiness matters
Signals Evolution and Trade-offs
- All non-React frameworks now use signals (including Angular)
- Developers understanding signal trade-offs better over time
- Still evolving with smaller innovations expected
- React appreciation growing after using signal-based frameworks
2025 Predictions
- No big technology leaps predicted
- Community may not accept major changes
- Focus on reflection and catching up with existing complexity
- Raw capabilities exist, need right combination in consumable form
- Good time to hunker down and get work done
- Boring year that sets up long-term exciting future
Key Technology Assessments
- HTMX: great option but won't take over world
- React: not necessarily more complex than other solutions
- Async/client-server interactions inherently complex
- Compilers can't solve all issues but help significantly
- React Compiler makes React code faster even if not React itself faster
