Top AI Product

So Cloudflare just dropped something wild. Steve Faulkner, one of their engineers, spent a single week rebuilding Next.js from scratch using AI — and the result is [vinext](https://blog.cloudflare.com/vinext/), a Vite-based alternative that honestly has no business being this good for a one-week project.

Here’s what caught my attention. vinext builds up to 4.4x faster than Next.js 16 and produces bundles that are 57% smaller (we’re talking 72.9 KB gzipped vs 168.9 KB). Those aren’t theoretical numbers either — they tested both frameworks against the same 33-route App Router application doing the same work. The secret sauce is that vinext sits on top of Vite 8 with Rolldown, which turns out to be an absurdly efficient foundation for a React meta-framework.

The AI angle is what really makes this story interesting. Almost every line of code was written by Claude through [OpenCode](https://github.com/sst/opencode) sessions — over 800 of them. The total cost? About $1,100 in API tokens. But before anyone screams “AI slop,” the codebase has 1,700+ Vitest tests, 380 Playwright E2E tests, full TypeScript type checking, and covers roughly 94% of the Next.js 16 API surface. That’s more test coverage than most human-written projects I’ve seen.

The [GitHub repo](https://github.com/cloudflare/vinext) is open source, and about 95% of vinext is pure Vite code — not Cloudflare-specific stuff. Right now it deploys to Cloudflare Workers, but they’re actively inviting other hosting providers to contribute adapters. There’s even a proof-of-concept Vercel adapter already.

Naturally, this [blew up on Hacker News](https://news.ycombinator.com/item?id=47142156) within hours. The discussion is exactly what you’d expect — a mix of genuine excitement, healthy skepticism about the “AI wrote it all” claim, and a lot of people reading between the lines about what this means for the Cloudflare-Vercel rivalry. A heavyweight infrastructure company rebuilding your framework in a week with AI is quite the power move, whether you love it or hate it.

vinext is still explicitly experimental and hasn’t been battle-tested at scale. But as a proof of concept for what’s possible when you combine a senior engineer’s architectural vision with AI’s ability to grind through implementation details, it’s pretty hard to ignore.