How we rebuilt Next.js with AI in one week | Hacker News Reader

How we rebuilt Next.js with AI in one week

blog.cloudflare.com

540 pointsghostwriternr236 comments

This is pretty fascinating and comes with some complicated AI-world incentives that I've been ruminating on lately. The better you document your work, the stronger contracts you define, the easier it is for someone to clone your work. I wouldn't be surprised if we end up seeing open source commercial work bend towards the SQLite model (open core, private tests). There's no way Cloudflare could have pulled this off without next's very own tests.

Speaking more about the framework itself, the only real conclusion I have here is that I feel server components are a misunderstood and under-utilized pattern and anyone attempting to simplify their DX is a win in my book.

Next is very complex, largely because it has incrementally grown and kept somewhat backwards compatible. A framework that starts from the current API surface and grows can be more malleable and make some tough decisions here at the outset.

Crazy to see it's already being run on a .gov domain[0]. TTFGOV as a new adoption metric?

[0] https://www.cio.gov/

> The better you document your work, the stronger contracts you define, the easier it is for someone to clone your work.

Well said; this is my thinking as well. One person or organization can do the hard work of testing multiple approaches to the API, establishing and revising best practices, and developing an ecosystem. Then once things are fairly stable and well-understood, another person can just yoink it.

I have little empathy for Vercel, and here they're kind of being hoist by their own petard of inducing frustration in people who don't use their hosting; but I'm concerned about how smaller-scale projects (including copyleft ones) will be laundered and extinguished.

> Then once things are fairly stable and well-understood, another person can just yoink it.

That transparency & availability for community contributions or forks is the point of open-source.

If you're only using open-source as marketing because you're bad at marketing, then you should probably go closed source & find a non-technical business partner.

Whoever "yoinks" the package runs into the same problem because they now have to build credibility somehow to actually profit from it.

Established corporations will be doing yoinking, with a pre-existing credibility. There's a huge incentive to offer these copied services for cents on the dollar, as a way to kill the competition.

Credibility doesn't transfer easily.

Anyone who yoinks an open-source package still has to present an argument about why their offering is better than the original maintainer's.

When push comes to shove, cheapness and hype can shove.

I agree, but it's historically been as simple as "you already use AWS for everything else". For example, ElasticSearch.

It'll be interesting to see if this happens at a service level too. Like how lots of companies offer an S3 compatible API, will companies start offering similar services and building a compatibility layer over the top as an easy way to for customers to transition? You could use the existing service as a test suite to check your compatibility API behaves the same as the original product.

> There's no way Cloudflare could have pulled this off without next's very own tests.

I'm very uncovinced. History showed us very complex systems reverse engineered without access to the source code. With access to the source code, coupled with the rapid iteration of AI, I don't see any real moat here; at best a slight delay.

Source code is one thing; tests covering the codebase are another.

And if you just copy the source code or translate it one-to-one into a new language, rather than make a behavioral copy, there will be copyright issues.

> there will be copyright issues

Next.js is MIT-licensed. Cloudflare's rewrite is... also MIT licensed...

Of course. I'm referring to rewrites of other software; you can easily launder GPLed code this way, for example.

The tests are absolutely essential, otherwise there's no signal to guide the LLM towards correct behavior and hallucinations accumulate until any hope of forward progress collapses.

Obviously the signal is comparison against the behavior of the original.

I have tried to post this here but it has not got traction.

I have a demonstrated process here on my blog (all hand written without AI).

This bit about how to brute force decompilation: https://reorchestrate.com/posts/your-binary-is-no-longer-saf...

And this about how to do the conversion and address the LLM hallucination problem: https://reorchestrate.com/posts/your-binary-is-no-longer-saf...

Yes, it is absolutely possible.

There was a recent post on here where the creator of Ladybird (Andreas Kling) translated a chunk of his novel browser from c++ to Rust in two weeks -- a feat he estimated would take him months: https://ladybird.org/posts/adopting-rust/

I, in my own way, have discovered that recent versions of Claude are extremely (as in, super-humanly) good at rewriting or porting. Apparently if recently released coding agents have a predefined target and a good test suite then you can basically tell them that you want X (well-defined target w/ good suite of tests) written in Y (the language/framework you want X written in but it isn't) -- and a week or two later you have a working version.

I have spent the last month wrapping my head around the idea that there is a class of tasks in software engineering that is now solved for not very much money at all. More or less every single aspirational idea I have ever had over the last 20 years or so I have begun emabarking on within the last two months.

I hear you.

I am curious, have you attempted to do this to any binary packed with commercial obfuscation/"virtualization" schemes (e.g. Orean's Themida/Code Virtualizer and VMProtect)?

No, I would need to find a binary to test on. I suspect it would produce horrible code at the decompiler layer but ultimately I would expect that function signatures are still relatively clean?

Its scary - once you get the differential testing harness set up it seems to be just a matter of time/tokens for it to stubbornly work through it.

> I wouldn't be surprised if we end up seeing open source commercial work bend towards the SQLite model (open core, private tests).

Wouldn't this just mean that actual open source is the tests? or spec? or ... The artifact which acts as seed for the program, what ever that ends up being?

The point is that artifact is not open.

I'm not sure about this. LLMs can extract both documentation and tests from bare source code. That said I think you're correct that having an existing quality test suite to run against is a huge help.

Man, I love Next ... but I also love Vite ... and I hate the Next team, because they focus on fancy new features for 0.1% of their users, at the complete expense of the other 99.9% of the Next community (who they basically ignore).

This gives someone like me everything we want. Better performance is something the Next community has been begging for for years: the Next team ignored them, but not the Cloudflare team. Meanwhile Vite is a better core layer than the garbage the Next people use, but you still get the full Next functionality.

I wish Cloudflare the best of luck with this fork: I hope it succeeds and gets proven so I can use it at my company!

What is it you love about Next that isn’t tied to Vercel and isn’t available elsewhere? I love Next too but I find the value is inextricably linked to Vercel. I can’t imagine choosing to use Next if I’m not choosing it for Vercel’s fancy stuff.

React server components are dope. Server actions are dangerous but powerful. No one has a more mature implementation of either of these than Next.

Of course no one has a more mature implementation of it than Next. The Next.js team designed it themselves!

I still just prefer having a more clear separation of concerns with API routes instead of using server components. I want my frameworks to be way less fancy than what Next is pushing out these days. I get the feeling we're dealing with the consequences of Vercel employees needing to justify their promotions.

It is hilarious to me that the industry has reinvented serving HTML to clients, but with many intermediate steps, and this is heralded as groundbreaking.

You'd almost wonder if there wouldn't be more to it.

React and RSC are not dope they are a kludge and the only reason you’re blind to that fact is because you’re React brained and have no experience with modern alternatives that are actually good like SvelteKit or SolidStart.

at my job we have some 7+ year old nextjs apps that don't receive new features but still do their jobs perfectly fine, and they keep changing random shit around for no reason, we've had to waste time on multiple refactors already for major nextjs version bumps once the older ones are no longer supported

Is there any front end framework that doesn't do this? I dropped out of the front end years ago, and it seems to just get worse every year with a profusion of confusion. Doesn't anyone yearn for back when we didn't have to build the front end at all?? Just emit some HTML and serve up some JS files from the backend, and everything just flows from there?

Someone go make an AI rewrite of Apache+Mod-PHP and sell it to zoomers as the hip new thing already please

> Is there any front end framework that doesn't do this?

React, if you are judicious about what additional packages you use on top of it.

> I dropped out of the front end years ago, and it seems to just get worse every year with a profusion of confusion.

This has actually gotten somewhat better in recent years starting with esbuild which made it possible to use a simple single-binary tool for bundling.

What do you use?!

I know everyone loves to hate Angular but it is in a really good place at the moment. If you don't need SSR and just want to build an SPA, Angular is the way to go imho.

Is there any reason to keep upgrading if the apps keep doing their jobs perfectly fine? Pull in a stable version of the framework and the associated docs and stay there.

Security.

For example: https://nextjs.org/blog/CVE-2025-66478

Sounds like Tapestry. Had a friend who loved it but he stopped talking about it after the 4th major architecture shift.

I'm moderately hopeful that LLMs will help here because they lack the human motivations to needlessly mess around with stuff and over-complicate things.

I mean you don't want really want to use javascript for the backend anyway... What's the problem with just using vite and any backend of your choosing?

Next is the worst framework I’ve ever used next to rails. It’s pure overhead for most apps.

Try mvc-router, see here: https://github.com/wisercoder/mvc-router/tree/master/DemoApp...

React was originally meant to be the 'V' in MVC. You can still use it that way and React becomes very simple when you only use it for UI. Why do data fetching in a React component?

I built something similar actually: https://starfx.bower.sh

Rails 8 is surprisingly good nowadays. It absolutely still has its share of problems (e.g. Bundler being slow, the frontend story being crappy without Inertia, lack of types which is a biggie, memory) but it is still a fantastic framework imo.

Why Inertia.js? I quite enjoy not using JS heavy frontends in Rails by leaning on Turbo and light Stimulus JS controllers where needed. My experience going hard into Vue+Rails was full of pain and I've rediscovered why server first makes everything easier to reason about instead of duplicating tons of logic + dealing with constant async issues (particularly around automated testing and complex data loading).

Inertia because it’s a plug-in replacement for ruby html templating aka erb. Try it out, it’s basically the same stuff you get from erb, without the need for Turbo’s web sockets. You get server side rendering, all the great BE stuff like server side validation, but no SPA headache.

I find the best DX with Adonis/nodejs and typescript.

Somebody should port uv to Ruby :/

It's called rv: https://github.com/spinel-coop/rv

It’s unbelievably terrible. I don’t understand its success at all, as much as I’ve tried.

The basic premise of Next is good, but it definitely has more overhead that in should, has odd "middleware", and is very hard to optimize. I view this mostly as a React problem though since any page requires full hydration and ships everything to the client. RSCs are... not my favorite for sure.

I too have been very frustrated by this, and I made an "Astro for dynamic sites" TypeScript framework called Hyperspan ( https://www.hyperspan.dev ) that aims to fill the gap in the JS ecosystem for a modern fully dynamic option that, similar to Astro, makes dynamic islands easy. I have enjoyed using it in all my own projects. Check it out if you want.

RSC by design does not ship everything to the client. That's one of its basic premises. It ships markup, composed in client interactivity, but you can shed a lot of the code required curate that markup.

I obviously meant traditional React components, not RSC. RSC can eliminate some client code, but they can be very awkward to use in practice, and lines between server and client get blurry really fast. The mental model is difficult for many to fully grok. I say this as someone who has lead engineering teams with folks of varying skill levels. RSCs are not worth the extra complexity and mental overhead they bring.

If Rails is considered a worse framework, then I'm pretty much speechless. Not everything has to be about performance. Security is a thing too.

Rails powers nearly 15 percent of the US e-commerce. I love it. Any time I have to use another framework it feels like a huge downgrade. Rails has so many things that make it nice to use

Agreed 100%. I shifted to Phoenix/Elixir years ago but I still love Rails for all the sensible defaults it provides.

How is the elixir job market these days?

You can find specialized roles than Rails and usually more experienced (understandably) but most companies would be really open if you told them you have a Rails background but want to learn Elixir.

The pool size is less, but the pay is more (depending on your demographics, experience, etc) in my personal experience.

But, honestly I chose it not for the market, it's just a better programming language to build stuff, period.

Elixir is nice but the console feels like the Stone Age by comparison. That drives me a little crazy

I have the opposite opinion - give me stone age over the colored christmas lights anyday :)

It's terribly painful to go and have to import everything you want to use when you want to use the Elixir console. That alone makes it not worth it to me.

I miss Rails so much when working with any of the top JS frameworks.

Every time I run into an issue that Rails had a standardized solution for a decade ago just proves that most of the JS world spends their days metaphorically digging holes with sharp sticks, rather than using the appropriate tool.

But the industry values overpaying stick-diggers over results, therefore I gotta play along…

Shilling a bit but maybe check out wasp.sh, we are conceptually very similar to "Rails for JS"! Just don't tell Claude to completely copy us hehe (pls)

Nothing personal and I wish you the best of luck with wasp.sh, but the constant churn of new libraries that will revolutionize JS development, but are never quite finished and are eventually abandoned in a semi functional state is exhausting and exactly the main issue I have with the JS ecosystem in general.

At this point, I'm convinced there's a secret global conspiracy to prank JS developers. For example 1 person maintaining 3 similar-but-distinct decimal libraries for Javascript, or the top 3 PDF processing libraries silently producing blank outputs.

I get that people like but I really cannot stand most of the decisions it made. The worst of all is auto imports which every language has rejected: https://bower.sh/on-autoloading

Yeah, Vercel should have done this with NextJS a while ago. There is a reason why quite literally every other framework uses Vite because it amazing, easy to use, and easy to extend.

Everything just becomes a plugin.

It may be sacrilege to bring it into this conversation, but I've spent the last year building a fairly large community site in Nuxt, vite has been wonderful, though I prefer vue over react. I am a little annoyed I paid for NuxtUI Pro like 3 months before it became free, but whatever.

I don't get it, honest question (like the others), what exactly do you love?

You think you'll get better long-term support from an experiment that a single engineer did in his spare time?

Isn't that how Linux started?

No, not at all

For all I know Linux started a hobby project from one person.

So entertain me if you disagree.

Actually, that's exactly how it started: read Linus Torvalds' "Just for Fun".

Was Linux owned by a large company? Was the maintainer getting paychecks from that company? Was it profit motivated? Was it released as an AI experiment?

If the similarity is "they are both open source projects" then so are about a million others. 99.99% of them don't get any traction beyond the first week.

The similarity is to,

> You think you'll get better long-term support from an experiment that a single engineer did in his spare time?

Linus started it as an experiment. That's a single engineer doing it on his spare time.

Do you think Linux doesn't do long-term support right?

The one changing the goal post is you.

https://github.com/cloudflare/vinext It is MIT licensed. It can be used and maintained by anyone.

If it'll get adoption like Linux did, that's different. But the base is there.

what if we all move to vinext? I'm asking claude to migrate us in a git worktree using a team of agent, installed the vinext skill to help with that, it did it in 10min

also why do you need support? agents are the support

Weird, I hate Next and I love Vite. We have a big (I mean _really_ big) production app that runs on Next.js at work and it's the slowest thing I've ever worked on. I had to upgrade my machine to an M4 Pro just to get local dev compile times down from 5-8 minutes to ~30-60 seconds per route. And my hot refreshes are down from ~15-20 seconds to 5-10. It's _bad_. All the Next.js team does is give you the run-around and link to their docs and say here, try these steps, you're probably doing something wrong, etc. Nope. The framework is just slow. They use simple toy apps to demo how fast it is, but nobody tells you how slow it is at scale.

Damn, that's bad. You can compile C++ faster than that!

If you are using webpack, see if you can make the switch to turbopack. It cut my build times from ~1 minute to 15 seconds, incremental builds are down from 10 seconds to 2. Memory usage is down a ton as well. But if you rely on webpack plugins this may not be an option for you.

I've been pushing our apps at work in this direction and it's definitely worth it.

If you're relying on webpack plugins heavily, I'd consider that a liability either way. It's going to seriously hamper your portability to other frameworks and build tools and even new versions of the current ones.

You can easily run turbopack for development / preview environments and webpack for production(-like) ones btw. as long as you don't rely on custom magic.

This is EXACTLY how I've felt. We've said it out loud multiple times but I think we never did it together.

I just wrote an open letter to fix Next.js. The goal is to compile a list of needs for whoever decides to fix those in any way possible...

https://please-fix-next.com/

I wonder to what extent you should say you "rebuilt" something when the most basic hello world example doesn't work. And I wonder to what extent it makes sense to call it "from scratch" if you inherit a battle tested extensive test suite from the thing you're rebuilding, and the thing you're rebuilding is part of the training data.

Here's the first paragraph of Harry Potter and the philosopher's stone. I rewrote it from scratch, apparently:

Mr. and Mrs. Dursley, of number four, Privet Drive, were proud to say that they were perfectly normal, thank you very much. They were the last people you’d expect to be involved in anything strange or mysterious, because they just didn’t hold with such nonsense. Mr. Dursley was the director of a firm called Grunnings, which made drills. He was a big, beefy man with hardly any neck, although he did have a very large mustache.

is it you, rauchg?

In these AI hype cycles "works" is not a requirement for announcing AI did something successfully.

The web browser from cursor didn't compile

The c compiler from anthropic couldn't build stdio

And now, the Next.JS clone from cloudflare can't do a hello world.

I find it interesting that they bought Astro (https://blog.cloudflare.com/astro-joins-cloudflare/), which from my definitely-not-a-frontend-person perspective seems to tackle a similar problem to Next. A month ago.

If it is so cheap to make something that they recommend using (rather than a proof of concept), why buy Astro (presumably it was more expensive than the token cost of this clone?).

One conclusion is that, at the organisational level, it still makes sense to hire the “vision” behind the framework, rather than just clone it. Alternatively, maybe AI has improved that much in 1 month!

I think they just want steer users/developers to CF products, maybe not? It is interesting to see the two platforms. I've moved to svelte, never been a frontend person either but kind of enjoying it actually.

Astro is a different paradigm. Acquiring Astro gives Cloudflare influence over a very valuable class of website, in the same way Vercel has over a different class from their ownership of Next.js. Astro is a much better fit for Cloudflare. Next.js is very popular and god awful to run outside of Vercel, Cloudflare aren’t creating a better next.js, they’re just trying to make it so their customers can move Next.js websites from Vercel to Cloudflare. Realistically, anyone moving their next.js site to Cloudflare is going to end up migrating to Astro eventually.

Can you talk more about this? What’s wrong with cloudflare pages plus Nextjs? Why do you need Astro?

Thanks

> which from my definitely-not-a-frontend-person perspective seems to tackle a similar problem to Next.

It does not. Astro is more for static sites not dynamic web apps.

That used to be the case 3-4 years ago. Today Astro is very much a serious contender for dynamic web apps.

I tried it about 6 months ago for something I had to redo in NextJS afterwards, it is really not built for those sorts of web apps, even today.

What features are missing?

Unlike Astro, this looks more like a low effort experiment while making fun of a competitor than a project intended for any serious production use.

Maybe I'm wrong. We'll see what happens a couple of years from now.

I'm very patient with the ai-led porting projects since they're revealed with a big engagement splash on social media. Could it be durable? sure but I doubt anyone is in that much of a rush to migrate to a project built in a week either.

I view it as a long-overdue exit ramp for maintainers of Next.js-based webapps to extricate themselves from its overly-opinionated and unnecessarily-tightly-coupled build tooling. Being stuck on webpack/rspack and unable to leverage vite has been a huge downside to Next.js. It's a symptom of Vercel's economic incentives. This project fixes it in one fell swoop. I predict it hurts Vercel but saves Next.js.

You buy that concern to hire the people. The stack is already free.

Astro has "server islands" which rely on a backend server running somewhere. If 90% of the page is static but you need some interactivity for the remaining 10%, then Astro is a good fit, as that's what makes it different than other purely static site generators. Unlike Next.js, it's also not tied to React but framework-agnostic.

Anyways, that's why it's a good fit for Cloudflare: that backend needs to be run somewhere and Astro is big enough to have some sort of a userbase behind them that Cloudflare can advertise its service to. Think of it more as a targeted ad than a real acquisition because they're super interested in the technology behind it. If that were the case, they could've just forked it instead of acquiring it.

From Astro's perspective, they're (presumably) getting more money than they ever did working on a completely open source tool with zero paywalls, so it's a win-win for both sides that Cloudflare couldn't get from their vibe-coded project nobody's using at the moment.

Astro isn’t solving the same surface as next. Astro is great for static sites with some dynamic behavior. The same could be said about next depending on how you write your code, but next can also be used for highly dynamic websites. Using Astro for highly dynamic websites is like jamming a square peg into a round hole.

We use Astro for our internal dev documentation/design system and it’s awesome for that.

Astro is not a static site builder.

But presumably if you could do this for Next it would be at least as easy for Astro?

This is probably the most interesting AI experiment I've seen yet. Looking through the codebase has me wondering where all the code is. I don't know if anyone has had the displeasure of going through the next.js codebase, but I estimate it's at least two orders of magnitude more code than this reimplementation. Which makes me wonder, does it actually handle the edge cases or does it just pass the tests.

Like compare the two form implementations for example. Vinext is a completely different implementation compared to what the Next.js version does. Is their behaviour actually the same? The rewrite looks incredibly naive.

https://github.com/vercel/next.js/blob/b8cbaad24ca66ec673a7b...

https://github.com/cloudflare/vinext/blob/main/packages/vine...

Either way, pretty impressive.

It is the most passive aggressive thing I’ve ever seen. Cloudflare team had issues with the Next team? And they responded with ‘we can do your whole product with an intern and AI’, lol.

Woah.

Recreate your competitor's product with an intern + AI over a weekend. Hilarious!

I hope this becomes common practice. It might even work as an interview question for hiring new candidates.

So you think a reasonable interview question would involve spending hundreds to thousands of dollars on tokens?

Yeah, I'm not even sure what this is, or whether or not this is even intended to be a serious project. It just comes across to me as deeply unprofessional. I say this as someone who doesn't even like Vercel and has their own gripes with that bizarrely run company.

The docs say that none of the code has been reviewed or tested properly, so how serious is this for people to run in a production setup where many companies are going are not going to be super enthused that their production environment has been 'vibecoded' in a week? It just reads to me like a not-so-subtle middle finger to the Vercel guys.

On a side note, I find it extremely weird that the current AI era of software development is turning the state of the entire field into a reenactment of Lord of the Flies. Bizarre behavior all around from people who should know better.

I imagine offloading a lot of the heavy lifting to Vite helps cut down on the code size.

This. Whole thing struck me as basically an advertisement for Vite. 99% of the base functionality is probably already there, written by humans.

"Use our proprietary SaaS and you too can approximate Next.js in 1/100 as much code using a bit of chicken wire and an LLM".

Whole thing sounded too good to be true, and it was.

(to be transparent - I'm a Cloudflare engineer)

The behavior isn't entirely the same and reaching 100% parity is a non-goal, but there are a few things to note.

This is still a very early implementation and there are undoubtedly issues with the implementation that weren't covered in next's original test suite (and thus not inherited) while not being obvious enough to pop up with all the apps we've tried so far.

As for why it's so much smaller, by building on top of Vite and their react + rsc plugins there is a whole lot of code that we don't need to write. That's where a significant portion of the LOC difference comes from.

Yeah I'm curious about all the routing edge cases, form actions, server functions etc, since that is where most of the complexity of the app router comes from. Does it encrypt captured values inside closures sent to the client? Stuff like that.

Why is parity a non-goal? The blog post states:

> The result, vinext (pronounced "vee-next"), is a drop-in replacement for Next.js

"Drop-in" in my mind means I can swap the next dependency for the vinext dependency and my app will function the same. If the reality is that I have to spend hours or days debugging obscure edge cases that appear in vinext, I wouldn't exactly call that a drop-in replacement. I understand that this is an early version and that it doesn't have parity yet, but why state that it is a non-goal? For many of us, that makes vinext a non-choice, unless we choose to develop for vinext from the beginning.

Furthermore, if you're making a tool that looks almost like a well-known and well-documented tool, but not quite, how is gen AI going to be able to deal with the edge cases and vinext-specific quirks?

Changing the definition of drop-in definitely has me concerned and makes me not take this any seriously than other projects open-sourced by Cloudflare, particularly the ones focused on more critical parts of their systems – e.g. pingora and ecdysis.

Makes me wonder how the hell all the tests passed?!

I ported the number theory parts of sagemath in typescript, I thought that was pretty interesting as well xD https://github.com/zksecurity/sagemath-ts

Nextjs had remote code execution vulnerabilities because of how they implemented react server side. I am not touching an AI version without waiting for a while.

Thank you. This is the part that shocks me the most. I was always wary of Next.js for this exact reason (in fact, I refused to use it for personal projects before the RCE because I was scared that I would make a mistake and leak server-side data to the client.

Bugs like this are easy to happen and even easier to miss if you’re generating thousands of lines of code with AI.

That was a React vulnerability, not a Next one.

It was a vulnerability that only could exist due to the incestuous relationship between React and Vercel. It was something Vercel has been trying to heavily push into React for years (which is why they hired previous react core team members).

I'm deeply skeptical of the "X reimplemented and it was super easy" thing.

The devil is in the detail.

So many edge cases unlikely to be there.

So many details or fine details unlikely to be there.

Years of bug fixes.

If it is literally a drop in replacement and it passes all the tests, and you're replicating something with and extremely thorough test suite, then sure I'll give you the benefit of the doubt.

Otherwise, I don't believe people "rebuilt X product in a week".

I don't necessarily buy it either, but TFA talks about the test suite. They basically pulled 2k unit tests and 400 E2E tests from Next and made sure they all passed.

Even then I still don't think passing all the tests means you duplicated something. That would be a naive understanding of the reality of tests.

You think Next.js writes a test for every bug?

I defer to my first 5 words above.

God speed to the poor souls that have to make it actually work in the long run:

"I can say that with some authority. Yes, I'm the one who wrote most of this project, but I'm also the director in charge of the entire Cloudflare Workers org, almost 80+ people at this point. I'm not just an IC engineer who has to find time to justify continuing to work on this. I can literally put people on it, and I've already been talking to the team about how to do exactly that."

https://github.com/cloudflare/vinext/issues/21#issuecomment-...

> Most abstractions in software exist because humans need help. We couldn't hold the whole system in our heads, so we built layers to manage the complexity for us.

Kind of a sloppy statement, but I don't think it's accurate to say abstraction or layering exists in software just because humans need help comprehending it. Abstractions often exist to capture the essence of some aspect of the real world, and to allow for software reuse. AIs will still find reusing software useful? Secondly, you equate "abstractions" with "layers" which aren't really the same thing. Layers are more about separation of concerns. Maybe it could be argued layering is a type of abstraction.

Here's what is buried a bit in the text:

--- start quote ---

Something like 95% of vinext is pure Vite. The routing, the module shims, the SSR pipeline, the RSC integration: none of it is Cloudflare-specific.

--- end quote ---

The real achievement is human-built Vite (and it is an amazing project).

Since Next.js's API surface and capabilities are known, this is actually quite a good use of AI: re-implement some functionality using a different framework/language/approach. They work rather well with that.

> The real achievement is human-built Vite (and it is an amazing project).

From TFA:

Vite is the build tool used by most of the front-end ecosystem outside of Next.js, powering frameworks like Astro, SvelteKit, Nuxt, and Remix

Are you saying those frameworks aren't impressive because they are also powered by Vite?

Also from TFA: A project like this would normally take a team of engineers months, if not years. Several teams at various companies have attempted it, and the scope is just enormous. We tried once at Cloudflare! Two routers, 33+ module shims, server rendering pipelines, RSC streaming, file-system routing, middleware, caching, static export. There's a reason nobody has pulled it off.

That's the most important result of this experiment. They achieved something that they'd wanted to do but couldn't pull it off. Do you think they are lying?

> Are you saying those frameworks aren't impressive because they are also powered by Vite?

That is not what I'm saying

> That's the most important result of this experiment. They achieved something that they'd wanted to do but couldn't pull it off. Do you think they are lying?

Once again, that is very explicitly and very clearly not what I'm saying or thinking.

You could try actually reading and understanding what I wrote instead of responding to words in your head.

Feels a bit like the AI browser that in the end imported Servo for all difficult things.

I remember multiple people at HN saying "show me ONE example where AI was used to produce commercial-grade software" like a month ago. Cloudflare alone has posted a couple of examples recently, and yesterday Ladybird was ported to Rust using AI.

The most interesting aspect I see in all these examples is that extensive test suites make the work very straightforward. Maybe AI will produce a comeback of test-driven development.

Honestly, might try this in my next project. Just code the tests and have AI implement against it.

How is this production grade? The last few things CF posted with AI were outright lies or omitted large swaths of functionality.

If your stance is assuming we have an existing implementation of something in the training set, and we have a robust test harness already, and we have thousands of dollars to throw at tokens, and we're not at all concerned with "works" THEN this is viable then sure? But that doesn't seem to be what most boosters are saying.

Claudflare has mainly posted vibeslop with accompanying blog posts making wrong basic statements about the project.

I have done a similar thing with my web app (codeinput.com) and honestly wouldn’t touch this thing for even a fun project. My reason to migrate was two fold: nextjs simply didn’t work with Cloudflare workers and IBM carbon required ‘use client’ for every page which meant that no HTML was generated. Everything was client side.

Google Gemini, at the time, created an SSG solution which I had spent the next 3-4 months fixing bugs for. Consequently, I had to understand the whole SSG build step and all the wrong design decisions the AI made that resulted in the site getting a horrible core web vitals score. In the end, I just put the site behind a white “div” that disappear when the page finally loads. SSR is way more complex than it sounds.

This project (along with the quantum post) is quite concerning. It’s not clear why Cloudflare has decided to take this direction. If you want to know why LLMs are completely unable to produce something even close to NextJS, a better solution would have been to ask the LLM to fix the opennext adapter rather than building a new framework from scratch.

> The whole thing cost about $1,100 in tokens.

I like this is called out.

> We also want to acknowledge the Next.js team. They've spent years building a framework that raised the bar for what React development could look like. The fact that their API surface is so well-documented and their test suite so comprehensive is a big part of what made this project possible.

Hi next.js devs, we like to acknowledge the effort you put for writing good tests so we were able to rip it off. You know claude already has next's entire source code in it's training data?

Clownflare engineers are in the chat now doing damage control and downvote opposition. Soon jgrahamc will appear and whitewash the theft.

It's wild. If I need to steal someone's code, I just need to let an ai crawl it and it'll regurgitate the same thing. I can't wait for it to recreate marvel movies.

I think this framing is wrong. we have to learn to accept that it's not stealing. It's a new world where it's fair use and we don't know how to deal with it.

If we accept this then we can do something about it. Without it no one will heed us.

I understand your point, but at some point someone needs to think about morality.

If you or I copied and reimplemented nextjs in a better way, it doesn't feel as wrong. But when a large company does that and then brag about it, it's in poor taste.

Especially pointing out one developer and 1000 usd of tokens replacing the efforts of hundreds of talented developers. There's people on the other side of the screen.

The money points hurts yes. We were worth 200 per hour, and as of yesterday were compared to bring 2 dollars per hour, and going down.

The thing is feeling bad or complaining about this does nothing. What else then? Acceptance?

what's up rauchg?

"I'm also the director in charge of the entire Cloudflare Workers org, almost 80+ people at this point. I'm not just an IC engineer who has to find time to justify continuing to work on this. I can literally put people on it, and I've already been talking to the team about how to do exactly that." -Steve Faulkner https://github.com/cloudflare/vinext/issues/21#issuecomment-...

This is basically distillation for JavaScript frameworks. We're not creating new ones fast enough. So we'll just start distilling the ones we have.

In hindsight, a totally expected achievement given where models are and the high quality tests available, but wildly impressive all the same.

I don’t know what this means but it feels like yet another milestone moment.

The irony of this in the README:

- Node.js production server (vinext start) works for testing but is less complete than Workers deployment. Cloudflare Workers is the primary target.

The article say that "Next.js is well-specified." I... don't think this is actually true. It certainly has lots of documentation, but as has come up time and time again, there are tons of undocumented or poorly documented behaviors that have been the cause of consternation.

So I kinda wonder, did they just create the framework that Next.js claims to be but never has been? And is Next.js without the hidden stuff actually a good framework? Who knows.

The primary reason I moved away from NextJS is the constant stream of bugs. Never seen any framework or library perform as poorly as it past v0.

I don't want to argue if this is good/OK/useful, etc. I'm just wondering where this will go? Assuming it becomes so easy to copycat everything you found online, what would be the motive for anyone to do OSS? If it's trivial to copy, no one will ever care about maintaining any of that. They will just keep it private to themselves or the company they work for.

And the long-term consequence: we will all have to learn to build all of that ourselves. Yes, with the help of AI, but now the entire idea and architecture must be internal because there is nothing you can copy from.

I agree this might be helpful to an extent. We don't need another leftpad or isodd library. There are lots of things we should keep internal and do ourselves. And with AI, it becomes trivial. But if people try to play the whole OSS to their advantage, I think this can end up badly for everyone.

This boils down to one thing: If this is the future of OSS, give me one good reason for anyone to do anything open-source (or even source-open). And then show me from what those models will learn... The old stuff plus what you have internally? Have fun with that.

> In early benchmarks, it builds production apps up to 4x faster and produces client bundles up to 57% smaller.

> And we already have customers running it in production.

Wouldn't be like Claude to maybe forget to implement half the library, would it?

I guess they can call themselves Claudeflare now ;)

fwiw, I just tried running the agent-skill they provide for fun to migrate an app-router based next 15 site and the end result is it entirely failed to start.

Vite just hangs when running vinext dev, with no output in logs whatsoever beyond printing`vinext dev (Vite 7.3.1)`.

NextJS is bad enough, cannot imagine an Ai version

Cloudflare also lost my support because their support is among the worst, rep evn sneered (cannot update my WHOIS, still, after months of emails). Strongly recommend avoiding their platform. You will find that you lose more time & money to dealing with the issue of parity. God help you if you ever need support, almost every question in Discord goes unanswered as well.

The former CTO commented a lot here and said numerous times about emailing him with issues that support couldn't figure out. Maybe try emailing the new CTO?

nah, they already lost my business, it seems cultural, which we know is a hard ship to steer in a new direction, I'm not interested in trying that again

if said CTO happens upon this, my handle should show up in your systems if you do do as parent commenter suggests

i love how this disintermediates the next.js/vercel axis, which seems to be determined to make basically everything hard except for exactly what they want to do. as much as i love what vercel has done for open source in general (amazing stuff!) it is hard to interpret some of the stuff they do with next as anything other than vendor lock-in bs… the kind that i know is not in their hearts.

someone spent over 1000 dollars to replicate the functionality of Next.JS, even 1 dollar would seem too much somehow. I suppose that is me being overly retributive.

> it builds production apps up to 4x faster and produces client bundles up to 57% smaller.

I suppose that is you being overly retributive indeed.

coulda forked it for free

This tells me everything about recent cloudflare outages

All my homies hate Next.js

...and (again), hello world does not work [1]. the ai slop pr [2] absolutely butchers the fix. anyone foolish enough to switch to this is in for a rough time. details matter!

[1] https://github.com/cloudflare/vinext/issues/22

[2] https://github.com/cloudflare/vinext/pull/31/changes#r284987...

It's going to get increasingly difficult to sell software when there is no moat to replication. We're quickly reaching the point where you can just tell an agent "learn what this software does and then code it".

Everything Cloudflare does is half-baked, You don't do anything completely, properly and thoroughly - is that in your DNA or habit.

Now, you will not develop it further, it will face the same fate as OpenNext and adaptors.

OpenNext was SST. So was OpenAuth. All from same team that is behind OpenCode CLI.

Would LOVE a more in depth article about the discussions with Claude and how you created the agents to accomplish this! Would be very interested to know what you found the agents were really good at and what they could improve on.

I especially am interested in the initial discussions with Claude about scope, process, and general gameplan. I find that this initial discussion requires a lot of shaping and a lot of input, would love to see more insights into that process from your end!

Awesome stuff

Lots of hate for NextJS in here so im wondering what people use as an alternative framework...

Gatsby? I used to use that one until the updates basically ceased to exist.

Vite with <insert your favorite here> - looks good, but at initial glance seems to favor just pure speed for any other feature support like MDX, advanced SEO, etc.

Roll your own with React and webpack? Good luck, and you'll probably end up with something that looks like the others I've mentioned above.

Just surprised many comments are just stating complaints about Next and not providing any counter examples, its very un-HN.

This stood out to me as well. The level of discourse is incredibly low compared to other topics. Huge amount of "next/vercel sucks" without any constructive or deeper argument made.

There is a similar vibe on the Next.js subreddit, just enormous amounts of shallow negativity. Very strange.

(I'm not saying there aren't valid reasons to dislike the framework or company but the way it's expressed is consistently incredibly juvenile for some reason)

Eh, it's been rehashed over and over again. People experienced with Next.js and other frameworks don't need to read the same constructive, deeper arguments being made ad infinitum. It's like when people say microslop or micro$oft, everyone knows what they're talking about.

Astro, Nuxt, SvelteKit, SolidStart, React Router (prev. Remix), among others

Vite has plugins for MDX, SSR, etc. You can easily build your own framework if you want in a few hundred lines of glue code.

The criticisms of Next / React are so ubiquitous you can metaphorically gesture in their direction and most people know what's up. It's like wondering why someone said "ai slop" and didn't provide an expose on the quality of AI writing.

"from scratch"? No way. It's using Next.js test suites and the AI is almost certainly using Next.js code as part of the output.

I get the gist here but I hate the tone of these sorts of posts. Imagine being a NextJS developer, pouring your heart and soul into it day after day, knowing the codebase inside and out, and seeing some dude on the Cloudflare blog bragging about how he rewrote your project in a week using AI. It's tone deaf. It's not impressive.

The tool is hella useful. The messaging is ignorant. This should have been a "we built a tool to deploy NextJS on cloudflare natively" instead of this AI brag.

Tone deaf? It's the reality. Developers shouldn't bury their head under the sand. Chart your course accordingly.

> Rewrote your project

That project would die without user's adoption. Be appreciative. Nextjs is an open source project. What is it with HN that constantly praise the virtue of open source software, but downplay that fact the moment they don't like the outcome?

Tbf, stuff gets rewritten all the time.

asdf was the hot shit for quite a while, with people (myself included) invoking all kinds of shell arcanum to make it faster - then mise (née rtx) came out, and it was game over. Compatible with asdf’s ecosystem, but infinitely faster.

Poetry was incredibly popular, along with various other competitors, and then uv came out.

I get what you’re saying about the AI angle, because it’s somewhat different when a human takes your crown by dint of pure skill, but it’s gotta sting either way.

I partially agree with you, because Vercel put a ton of effort on this one. And the repo (and features list) is massive.

On the other hand, I do believe they drifted too much and stopped listening. As someone else said it - most of the fancy features are used by 1% of the projects, and everything else is buried in hacks and workarounds.

I also agree the tone (and especially the AI brag) is a bit too much. And at the same time it's honest - it doesn't need to be THAT hard and complex. Nor slow :)

That's why I've written an open letter to fix Next.js, to whomever wants to do it (Vercel, Cloudflare or anyone else). Because we have needs, and we cannot play this game anymore...

https://please-fix-next.com/ if you're interested

The most obnoxious man in the middle of the Internet rips off other people's code. Who knew?

Does anyone have experiences with the EU alternative bunny.net?

I’ve been using bunny.net for a couple of years, albeit just for DNS / CDN, and a static site. Zero complaints with those, though.

I've been using their cloud storage and CDN for a couple of years. No complaints. Super reliable.

I love Cloudflare but not a big fan of Next, I love Remix though :) But getting to make things work on Cloudflare is a pain, hopefully they will make it easier with OpenNext. On the other hand , maybe they can do something better at infrastrucute level , rather than make it easier, like a lot easier to bring your own JS flavor.

I would not want to be working on Cloudflare's Vercel partner team about now.. talking about a diplomacy-forward role

This traffic guided on-demand build thing, can it be powered by a file which has a standard format?

That’d make it vendor-independent - any vendor could create tooling to generate the traffic file, and you could opt to fetch it at build time or check it into your repo.

I am so over reinventing the wheel in tech, I feel as if that's 99% of where the money and time goes

Great to see. Could have use this last month when we migrated from OpenNext on CF to React Router 7

Just you wait, I will post how I rebuilt cloudflare with AI in one week

Do it?

Woahhh... This is direct threat to Vercel's ecosystem capture model. All those projects they hoovered up then re-aligned and soft-locked into their hosting business.

I'm here for it.

This is amazing. At Wasmer we are a bit concerned of Next.js and this replacement might come in handy!

I actually was thinking on creating something similar. Congrats to the Cloudflare team

- since you guys are experimenting that much, i have a favor to ask

- could you rewrite next and react actually without using a virtual dom at all and use a compiler like svelte instead?

Solid.js might be interesting to you.

There’s even a framework already: https://start.solidjs.com/

What the status on Vite Rolldown? Is it nearing production?

rolldown is in RC. it is open for private preview so they can detect and fix bugs.

Vite 8 is in beta, so you can try it out already. See https://voidzero.dev/posts/announcing-vite-8-beta

Rolldown is in RC, meaning no more breaking changes (except for experimental features). See https://voidzero.dev/posts/announcing-rolldown-rc

That's cool and all, but them investing $1,100 and two weeks of one engineer's time doesn't yet give me confidence that they're in this for the long haul. It'll be interesting to see how long the long tail of remaining issues will be (it doesn't sound like in another two weeks, they'll have pre-rendering and cache components working), but I'm definitely not adopting this any time soon.

Also looking forward to Vercel's version of https://matrix.org/blog/2026/01/28/matrix-on-cloudflare-work....

Either it works, or it doesn’t.

You’ll figure it out.

> The [next.js] developer experience is top-notch.

let me add my own unqualified statement to that: no.

> Next.js has invested heavily in Turbopack but if you want to deploy it to Cloudflare, Netlify, or AWS Lambda, you have to take that build output and reshape it into something the target platform can actually run.

it's almost as if vercel had some kind of financial incentive to gear this towards their own platform.

> reimplemented the Next.js API surface on Vite directly

a clown car screeches to a halt; several burnt-out-bored oracle vs google lawyers climb out and, weirdly, i am there for it

all in all, it's definitely a good example of something we couldn't have done for $1100 pre-llms, but: should we have? did somebody consult the lava lamps?

Shots fired, Vercel folks better hide!

It would have been cheaper to just ask Claude to port it to react-router :)

The result of these heists is that no one will publish test suites on the Internet in the future.

The tone of the blog post is upbeat. What are the consequences? Is the new performance expectation at Clownflare to "port" one framework per week? Do you have to generate at least 20 kLOC per week? Aren't you redundant right now?

God cloudflare's blog quality has fallen off a fuckin cliff ehh. Used to be so good now its just llm slop both content and actual writing.

The core network products seem to be having a run of downtime issues too. Maybe they should focus on their homework before going out to play with the AI kids.

This is interesting to my on both a technical level as well as a social-political level. I wonder what impact "AI-washing" will have on licensing for example

Yup. This was so jarring to read. Shame.

Gotta hand it to 'em though - posting this less than a month after the Matrix boondoggle certainly is, uh, audacious.

I'm actually surprised that this is the only hit I get when I search these comments for "matrix".

This is another example that good tests (e.g. Next.js's own test suite) are SO incredibly important to making the AI able to work on big projects autonomously with lower steering. So is a very domain-knowledgeable human in charge of steering.

Next-js team is a bunch of inexperienced teens who like good looking UIs and their lousy platform to upsell their services. Very glad to see this

Traffic-aware Pre-Rendering is smart

good job, now you have to support that mess that took 3774~ contributors to build.

have fun.

What else exactly would you expect for a competitor to do when trying to take a rival's market share?

I guess the thing here (which they admit in the post) is that they’re just porting it to Vite, which is the real champ of the story. The LLM basically worked as a translator instead of rebuilding the whole thing from scratch.

So maybe the project is sort of maintainable, as long as people maintain Vite.

for most folks -- just user Inertia + Django | Rails

again?

Nice. Now rewrite it in Rust.

THIS IS THE WAY.

Human in the loop, acting as an orchestrator.

Reply on news.ycombinator.com