In general Next.js has so many layers of abstraction that 99.9999% of projects don't need. And the ones that do are probably better off building a bespoke solution from lower level parts.
Next.js is easily the worst technology I've ever used.
I suppose the overly complicated ENV/.env loading hierarchy is (partly) needed because Windows doesn't (didn't?) have ENV vars. Same for inotify, port detection, thread management: *nix does it well, consistent ish. But when you want an interface or feature that works on both *nix and windows, in the same way, you'll end up with next.js alike piles of reinvented wheels and abstractions (that in the end are always leaking anyway)
Windows has pretty much everything you can dream of (although sometimes in the form of complete abominations), it's just that the people employed by Vercel don't give a shit about using native APIs well, and will map everything towards a UNIX-ish way of doing things.
If I had to create something that has a UI I'd just go with a bog standard server rendered multi page app, built using really boring technology as well. If you like Javascript and friends, go with Express. Nowadays you can run Typescript files directly and the built-in test runner is quite capable.
If a single page application makes sense, then go with vanilla React. For a highly interactive application that's potentially behind a log in anyway, you probably don't need React Server Components.
Things will get far worse before they get better. Right now, online courses such as the ones in PluralSight are pushing Next.js on virtually all courses related to React. I have no idea what ill-advised train of thought resulted in this sad state of affairs but here we are.
Windows' command prompt requires two separate invocations:
set KEY=value
./myApp
$env:KEY='value'
./myApp
[System.Environment]::SetEnvironmentVariable('KEY', 'value')
./myApp
I recently rewrote my auth to use better-auth (as a separate service), which has allowed me to start moving entirely off Next.js (looking at either React Router 7 or Tanstack Router).
Back when I started, Next.js made server side rendering incredibly easy, but it turns out I didn't need it. My marketing site is entirely static, and my app is entirely client rendered.
That is my opinion as well. Things like SSR are forced onto users with a very smooth onboarding, but I'm concerned that in practical terms this perceived smoothness can only persist if the likes of us pay the likes of Vercel for hosting our work.
In some degree I feel the whole React ecosystem might have ended up being captured by a corporation. Hopefully it wasn't. Let's see.
The reason for this, IT had contracted for a content management system from a Microsoft shop, because the CIO was a former Accenture/Avanade consultant. But the brochure-ware website had already been contracted to some random NYC-based web firm, but the CIO didn't want multiple usernames/passwords, so after the WordPress site hand been build, they hired the SharePoint consultants to build out the CMS that the employees would use, but it still didn't hook up to wordpress, so then it became another contractor's job (me) to join the two.
I had worked on Word Press, I even had a few decently popular plugins, but I had never seen the absolute hellscape that was SharePoint before. I wrote a codegen tool that would read the WSDL and create a library with all the classes and calls needed to use it without any SharePoint experience, and wrote some simple ETLs for the handful of "buckets". It was a 2-3 month long journey, but those libraries and my code are still in place today, where they still use wordpress for front-end, and sharepoint as backend (or at least did in 2022 still, the last I talked to anyone still working there).
Open up vercel, point it at my repo, and environment variable, it starts building and… build error. I search up the strange error log on the next issues page, and find a single issue from 3 years ago with one upvote and no response or resolution.
So I threw it on a VPS and just built it with whatever the “prod build” command is, and it totally worked fine.
So in my limited anecdotal experience, hosting it on vercel won’t save you either lol
As long as I can remember in my career, Windows had environment variables. So that's at least 25 years. It's both available to view/edit in the GUI and at the prompt.
By comparison, DIY SSR with Express takes a few days to get working and has run quietly for multiple projects for years on end.
And I have some use cases where I want to have a headless crm/api “hidden” behind the front end. So in these cases using next as a backend proxy works well for me.
I'm sure commercial incentives would lead issues that affect paying (hosted) customers to have better resolutions than those self-hosting, but that's not enough to explain this level of pain, especially not in issues that would affect paying customers just as much.
After a week of futzing with it I just threw up my hands and said 'no can do'. I couldn't untangle the spaghetti JS and piles of libraries. 'Compiling' would complete and if you looked at the output it was clearly missing tons of bits but never threw an error. Just tons of weirdness from the toolchain to the deployment platform.
It surely was a development platform, but wasn't supposed to be one.
I also had the impression they would probably follow the Vercel style, framework as a business model, with it being sold to Shopify.I don't really know where it's all going, but it is not the sort of thing I would tie myself to.
I don't know much about the nextjs and whether it was open like sveltekit currently is.
To me, nextjs (I think) was always meant to favour vercel but sveltekit has a rich history of managing multiple adapters.
Now, that being said there are still some chances of a rugpull that might happen but if that ever happens, I am staying on the last sveltekit that worked with cf and other cloud providers.
Pocketbase was the ONLY good thing about this journey. Everything else sucked just so terribly.
Infinite complexity everywhere, breaking changes CONSTANTLY, impenetrable documentation everywhere.
It is just so, so awful. If we rewound the last five years of FE trends and instead focused on teaching the stuff that existed at the time properly, we'd be in a much better position.
I've also built a very complex React frontend (few thousand users, pretty heavy visual computation required in many places). And while I don't particularly like React either, Next.js was even worse.
And lastly, built a CMS in Go, with vanilla JS. And while the DX sometimes feels lacking, I just can't help but feel that I actually know wtf is going to happen when I do something. Why is that so hard?
In React and Next.js I am STILL, AFTER SIX YEARS constantly guessing what might happen. Yes, I can fix just about anything these frameworks throw at me, thanks to all the experience I've gathered about their quirks, but it all just feels to messy and badly designed.
In Go, the last time I guessed what might happen was in the first six months of learning it. No surprises since. Codebases from years ago are still rock-solid.
Why can't we do this at the frontend, goddammit?
What's the story here? I assume this group was chosen for a reason and didn't meet expectations.
They paid them on the strength of seeing it working, but then the consulting group basically ghosted when the customer asked to adjust it to run on cheaper hosting (probably because they couldn't), then the site got shut off because the hosting was all in the consulting groups name and they stopped paying it. Digital Ocean nuked the database for non-payment and they lost tons and tons of manual work putting in data.
Tanstack seems to be following Next.js in that they’re just over complicating everything and their docs felt lacking for the most of their features.
It's pretty absurd to have such a broad range of web solutions, and think the same solution can cover everything.
If they had brought me in before hand I could have saved them a lot of work by asking the hard questions and reigning in the tech overspend.
I'm not so sure about that. We're seeing Next.js being pushed as the successor of create-react-app even in react.dev[1], which as a premise is kind of stupid. There is something wrong definitely going on.
Canvas instead of DOM -> :(
EDIT: Gave it another try and more issues appear, within seconds of using. The left side has a rendering bug where the selected areas are cut off sometimes, ctrl+zoom does not zoom the page as it does on all normal websites. I can still zoom via menu. Middle mouse open link in new tab doesn't work. Z layer bugs everywhere. I expect more the longer I'd look.
We do a 30-min tops exercise where you create a React project to show how to use useState and useEffect, etc. I help with whatever command they want to use and allow Google/ChatGPT.
More than half of the candidates had no idea how to use React without Next.js, and some argued it was impossible, even after I told them the opposite.
To many people, it's just basic logic: "everyone must want the latest React features, and the only way to get those is with Next, so everyone must want Next".
I switched to Astro from Next for most projects and haven't looked back. It's such a breath of fresh air to use.
The only "weakness" is that it doesn't have guard rails, so may not be great for larger teams with mixed experience.
This insanity of server side react introduces all kinds of unnecessary quirks.
Also, the VC-funded Vercel is of course purposely dumbing down Next.js, so that everyone pays them. Its a trap everyone should be aware of.
So, yeah, speaking in hindsight is really easy.
PS: no, the UNIX way is also shit, just in a different way.
That capture happened... two years ago? (Perhaps there's a good blog post there, if it doesn't exist already)
Sadly tan stacks releases a new version every other day and react router was complete 5 versions ago but cannot seem to keep changing the api to stay relevant in the never ending js relevancy ending war.
Vendor lock in. Magic leaky abstractions are great until you need to debug something a few layers down when the magic stops working.
> how else do you want framework development to happen?
Loosely affiliated open source efforts maybe. If that doesn't work, I would prefer to have none at all.
For me, lately, the interview question is "here's code that ChatGPT generated for (previous interview question as related to the role we're hiring for that we could do)", what's wrong with it? What do now? (ChatGPT may or may not have actually generated the code in question.)
While we would all like to retire to a cabin in the woods and be a carpenter, and for corporations not to exist, that seems unrealistic.
Magic leaky abstractions are orthogonal to vendor-lock in, and the source is open, so I'm not seeing the lock-in part. The "hey it's easier and cheaper to smash the deploy-to-vercel"-in, sure, but things cost money. Either to a developer, or to a company.
To be fair, this is partly on the kind of people who use it. E.g. if you're trying to build something that's intended to last for 10+ years but you don't think it's worth it to spend the 20 hours watching the Udemy course on Angular, then your technology is going to be a complete dumpster fire no matter which stack you choose.
>This insanity of server side react introduces all kinds of unnecessary quirks.
Im kind of confused by what you mean here. You can use PHP, Java, Ruby, etc. for the backend with Next.js. You can even use if for the SSR server if you want.
I guess you are actually talking about simply not doing server side rendering at all? Im just clarifying because I think that still constitutes using "React for the frontend." I mean, how could React be used for anything BUT the frontend?
The server side rendering is actually one of the objective goods Next.js offers (not necessarily that it handles the complexity of it well). I mean, if you dont need that, sure... that's one more reason to just use Vite. But the backend choice is irrelevant.
https://react.dev/learn/creating-a-react-app
It throws you straight at Next.js
How Powershell ever got popular is beyond me.
Next.js is essentially the reference and test bed impl.
Where people go wrong is thinking they need to default to the inherently complex niche feature of client hydration which is a niche optimization enabled by a quirk of web tech.
It is more like test on whether or not you can figure out random React minutiae (with Google/ChatGPT, if needed) when presented with a need. Which isn't a bad approximation for how well you will do at finding any random minutiae as needs present themselves. React-based development doesn't require much original thought — the vast majority of the job really is just figuring out the minutiae of your dependencies to fit your circumstantial need.
For fun, I asked ChatGPT for an answer and it gave a perfectly good one back without hesitation. Even if you had no idea what React was beyond knowing it is a library for developing web components, you should still be able to answer that particular question with ease.
When you're in a work meeting, do you just put ChatGPT up on one laptop and Claude on another and just sit back for 30 minutes to an hour?
Java doesn't offer isomorphic React SSR, but in most cases that is a questionable feature. Most SPAs don't need or want search-engine indexing or require instantaneous-seeming load times.
Objectively that sadness does not change reality however. At least within my own professional network no-one seems comfortable starting a new project using React today. Almost 100% of the paid front end work I've been involved with myself or discussed with others recently is now using alternatives - most often Vue though I've seen other choices at least seriously considered. I've even had a couple of recruiters I haven't worked with for years suddenly reappear desperately looking for someone to take on React work and openly admit it's because they are struggling to find anyone good who wants to go near it. All of this is a sharp contrast with the market of the early 2020s when React was clearly the preferred front end choice. And all of this is surely a direct response to the push to make React a full stack framework, the added complexity that has introduced, and the apparent capture of official React development by Vercel.
And while Spring has it's rough edges and quirks it is still an incredibly stable framework. Next, on the other hand, is a box of surprises that keeps on giving even when you think you saw it all.
Rich and Simon are incredibly important, but they're in it for Svelte and the community more so than a paycheck from Vercel. Tee has been doing most of the maintenance on SvelteKit currently funded by community donations. And this isn't counting other infrastructure like vite-plugin-svelte or the Svelte CLI which are entirely maintained by volunteers. I don't think Vercel funds a majority of the work on Svelte even if it might be close to it.
My point is that it's fishy how they push features that just so happen to be the value proposition of the only corporation that just so happens to be able to implement them.
Stuff costs money, sure. But I don't think it's that simple. Next and Vercel come from the same organization. I have no objection to a paid hosting solution making it operationally simpler. However when that same org has control over the free thing, they can make it even more easier (probably grammatical! who knows) that it would have "naturally" been.
I actually wrote exactly that blog post and did a conf talk on it earlier this year. I covered why the React team switched to directing users to use "frameworks" to build React apps, the development influences behind React Server Components, why the React docs didn't list tools like Vite as viable options until just a couple months ago, and various other related topics:
- https://blog.isquaredsoftware.com/2025/06/react-community-20...
- https://blog.isquaredsoftware.com/2025/06/presentations-reac...
Anyone who has ever maintained a semi complex set of bash invocations and pipes knows it's a fragile incantation that breaks anytime you look at it funny, or something in your chain produces unexpected output.
Powershell, while absolutely horrible to read and only slightly less horrible to write (hey look, proper auto completion instead of trying to cut on the 4th, wait no sorry 5th, ah fuck it's the 6th there's an invisible space) at least produces consistent and reproductible results.
No, your python script doesn't count, it makes me do a pip install requests. Oh, sorry, pip can't be used like that, gotta run apt instill python3-pip or my whole system breaks.
react-router if you just want a simple React frontend, write your backend in something else.
Framework-defined infrastructure is a seriously cool and interesting idea, but nowadays Next feels more like an Infrastructure-defined framework; the Vercel platform's architecture and design are why things in Next are the way they are. It was supposed to be "Vercel adapts to Next", but instead we got four different kinds of subtly different function runtimes. My usage dashboard says my two most-used things are "Fluid Active CPU" and "ISR Writes". I just pay them $20/mo and pray none of those usages go over 100% because I wouldn't have the first clue why I'm going over if it does.
Half the labels on there are Star Trek technobabel, which I would take the effort of learning, except I'm convinced they're all going to change with the next major release anyway. Partly because, I keep hoping & praying they will. I know a concerning number of former die-hard Zeit fans who've taken their projects and customers elsewhere. At the end of the day, if they were to ask me what they need to address in the next major release, I seriously do not know how to answer that question beside "practically every major and minor decision y'all have made since and including the App Router was the wrong one". How do you recover from that? Idk.
One of the factors is that web dev pushes for a complete separation of concerns, and thus allows frontend developers to specialize in front end development. Therefore it becomes far easier to hire someone to do frontend work with a webdev background than a win32/MFC background.
Number of applicants is also a big factor. There is far more demand for webdev than pure GUI programming. You can only hire people who show up, and if no one shows up then you need to scramble.
Frontend development is also by far the most expensive part of a project. In projects which use low-level native frameworks you are forced to hire a team for each target platform. Adopting technologies that implement GUIs with webpages running in a WebView allow projects to halve the cost. This is also why technologies like React Native shine.
Also, apps like Visual Studio Code prove that webview-based apps can be both nice to look at and be performant.
It's not capabilities. It's mainly the economics.
It's like not knowing how to write a for loop or how to access an object's property in JavaScript.
Then there came small web applications, and still no "front-end developers", since functionality could only work on the server.
It's only when AJAX was introduced in the mid 2000's that you could start to talk about "front-end developers".
By that time, win32 and MFC was old. We had Java, C# with .net framework, etc.
It's also dismissive of market forces, i.e. developers have to pay bills and therefore are easier to hire if they know the skillset that is in wide use.
I've never worked or interviewed a single senior that wanted to use Next.
It just adds a lot of complexity even if you don't explicitly opt in to it or need it.
I was part of a successful large project where we did our own SSR implementation, and we were always tinkering with it. It wasted a lot of time. Next.js "just worked". I've used Next with the pages router on two significant and complex projects and it was a great choice. I have no regrets choosing it.
The “static preview” it shows while it loads (for like 10-15 seconds!) is so much smoother and nicer to scroll around than the actual thing. On mobile, every third scroll attempt actually opens the right click context menu. It’s a stuttering mess on my high refresh rate phone. Nobody should ever make websites like this.
Or, if you insist, that Unix is inconsistent with how windows does it.
Which is what those wrappers and abstractions do: they expose a single api to e.g. detect file changes that works with inotify, readdirectorychanges, etc.