One Cloudflare engineer directed Claude AI across 800+ development sessions over seven days and produced vinext — a ground-up reimplementation of the Next.js API surface built on Vite 8 and Rolldown. Total cost in AI tokens: approximately $1,100. The result is not a wrapper or compatibility layer. It's a standalone framework that reads your existing app/, pages/, and next.config.js files without modification.
The Performance Gap
Vinext's benchmarks against Next.js 16 are striking:
- Production builds: 4.4x faster (1.67 seconds vs. 7.38 seconds)
- Client bundles: 57% smaller when gzipped (72.9 KB vs. 168.9 KB)
- Cold starts: Zero, when deployed to Cloudflare Workers
These gains come from swapping Next.js's Turbopack-based build pipeline for Vite 8's Rolldown bundler, which compiles Rust-native and avoids the overhead of Next.js's layered architecture.
API Coverage
Vinext currently covers 94% of the Next.js 16 API surface: App Router, Pages Router, React Server Components, server actions, streaming SSR, middleware, and Incremental Static Regeneration. A single vinext deploy command pushes to Cloudflare Workers, with native access to platform primitives like Durable Objects and KV.
The AI Development Story
The vast majority of vinext's code, tests, and documentation were AI-generated, though a human directed architecture decisions and priorities throughout. Cloudflare positioned this as a proof point for what one person working with AI can build against a framework backed by hundreds of engineers — and shipped it as an experimental project with known rough edges rather than a polished product launch.
Vinext is explicitly experimental and under heavy development, but it already has production users — including the U.S. government's CIO.gov website. For teams frustrated by Next.js build times or looking for tighter Cloudflare integration, it's worth watching even at this early stage.