That useState hook doesn't work the way you think it does. It's going to reset state every time the hook is invoked on rerender. It also isn't triggering a render, which is the entire point of the hook.
I think React is way overused but every time you see a blog post with "replace it with this pet mini project instead" I groan because you know the reaction is going to be "what about X feature", and of course the mini framework doesn't do it.
I've found the perfect balance to be Astro with Preact. You will inevitably have some piece of functionality that's complex (say, a contact form) and you can lean into Preact for that. But for the vast majority of a content-heavy site you can just use Astro and skip the client-side bulk entirely.
STANDARD DISCLAIMER ANY TIME I COMMENT ON FRONT END DEV: your project may be different. A blog or a shopping site and have extremely different requirements to a full Gmail-style web app. There does not need to be a one size fits all answer.
95% of apps people actually develop are really just CRUD and could easily get by without JavaScript on the front end at all. They certainly don't justify the layers of complicated state maintenance React and similar systems entail.
> 95% of apps people actually develop are really just CRUD and could easily get by without JavaScript on the front end at all.
Nearly every app has some version of CRUD, but you're conflating the persistence model with the interaction layer. Just because an app is CRUD doesn't mean it doesn't require complex, client-side interactions.
I'm building an EMR and there is a large portion that is CRUD, but that doesn't mean I don't rely on enormous forms or don't need client-side validation.
And not only, but "you could get by" is true of a lot of things. I could get by with full-page refreshes every time a form is saved or a chart is pulled, but that doesn't mean it's the best option or that the users won't notice it.
This argument has big "I seldom build web apps but have a lot of opinions on them" energy. Like, the idea that I, myself, have seen enough of the different projects and use cases for the web that I can unequivocally state something like "95% of apps don't need a front-end framework" reeks of ego. You just flat out haven't. The field is enormous.
It matters when you use an headless CMS from the MACH architecture enterprise culture, and they only support a specific SDK, which is mostly Next.js/React.
> You will inevitably have some piece of functionality that's complex (say, a contact form)
And this is why it pays to just learn React. It's never been easier. A contact form is not that complex. React won. The LLM can author it pretty well at this point, probably better (and cheaper) than at least 50% of FE devs. Treating a "contact form" as a potential fork in the road in 2026 is just sad.
It pays you, the developer, yes. It does not benefit the user.
> A contact form is not that complex.
Correct, my example was not a great one, just the first one that came to mind. Using a giant JS framework to do something as simple as a contact form should be seen as a failure of development culture.
> Treating a "contact form" as a potential fork in the road in 2026 is just sad.
Choosing React as a one size fits all because it suits you as the developer, with little to no consideration about what suits the user... that's what is sad in 2026. But it's where web development culture is. There should be no "React won". There is no need for there to be one winner. As developers we ought to be able to tailor solutions instead of be lazy.
SSG is solved for in React. The user is going to cope with a few extra KB of JavaScript just fine. Just like they cope when they watch Youtube or stream or w/e all day long. It's 2026. Data is cheap. The user really doesn't care until they hit your contact form that sucks because you tried to reinvent the wheel instead of just using React. Or when you need to branch out should your site ever do something actually interesting and this problem then gets even worse.
It's not hard anymore to get over the React learning curve. For a content heavy static site just use a good enough LLM and the right starter template and you'll never have to code, SSG and bundle size will just work, lighthouse will be all green. Your users do not care about the tech. The idea that React is the enemy of users sounds like a skill issue on your part. Good thing LLMs exist now, so you don't have to fret about the alleged perverse incentives. For a static site in 2026 a FE dev should be largely irrelevant. What could be more user friendly? Seems like big bad React has redemocratized the web right under your nose.
Data is cheap and fast, CPU time still isn't. I've profiled numerous web sites on slow Android devices and they absolutely crawl when parsing JS and giant JSON blobs required to hydrate a page that's one big React app for no real reason.
> It's not hard anymore to get over the React learning curve.
> The idea that React is the enemy of users sounds like a skill issue on your part.
You're contradicting yourself here. Is React incredibly simply to learn or is optimizing it for speed a "skill issue"? I'm not talking about myself here, I'm talking about the state of the industry. The web is full of slow-as-shit web sites that are held back by JS and JSON parsing. You don't need to convince me they aren't because I've seen it with my own eyes time and time again. As an industry it's become far more important to fart out basic functionality and move on than take any kind of pride in your output. It's sad and I'm not going to apologise for caring.
It's not a contradiction. You were/are unskilled. But it doesn't matter anymore. Just wanted to help you understand this is not a contradiction. You sound like someone who made your mind up about "the state of the industry" a long time ago. While this replay of debates settled a decade ago has been a nice, I must wish you a good day now. I don't mean to be too harsh. Ultimately you can use whatever tool you like, obviously. If it works for you then it works for you. Often that is what matters most.
You're being ridiculous and trying to being personal insult into a conversation. I have, numerous times, asserted that I am discussing the state of the industry as a whole. I don't understand why you've repeatedly tried to misinterpret this, it's in extremely bad faith.
Every time I see this kind of pet project I see whether they are addressing any actual issue of React or whether they don’t like React merely because it is popular. I especially look at whether they do state management or VDOM diffing differently from React. I find that if a project had done that, it is likely something the author was very proud of and was front-and-center. And there have been some interesting ideas in pet projects like using languages with macros so the build tool does part of the diffing. This one replaced VDOM with… nothing at all?
> Every time I see this kind of pet project I see whether they are addressing any actual issue of React
Then I think you're looking at those projects from the wrong angle. The point the author is making is that React is often used in situations where VDOM or complex state management aren't actually necessary. If all you're doing is e.g. validating a few form fields then neither is needed. But people use it anyway because React has become the default for frontend engineering no matter what.
Same, at least for smaller projects where I want some kinds of interactivity, but also a good static rendered page first and just the ease of writing, Astro+preact has been very nice.
I switched my small personal projects to preact in a (temporarily successful effort) to get one under 14kb after re-reading [0] a year or two back. So far haven't run into any limitations.
If you need React, then go ahead and use React, but I'd encourage people to try a simple s/react/preact/g on their websites. (Okay, not quite as simple as that, but even if you don't like vibe coding, this is something an LLM can be trusted with.)
Why should a contact form mean complex functionality? You can do it in plain HTML, even - is that complex already? Because validation? Like that (Google, it's you) address checking unable to place my house in the right village? Like phone number validation forcing you to add a 0 where no 0 is required to call me? And don't start me on streets and all that, or middle names, or dates or or or. Validations of form inputs are almost always a stupid yak shaving exercise, yet here we are.
Actually browsers already come with a minimal UI library. It's called HTML + CSS + JS.
There's actually no need to make any wrappers - you describe what you want in HTML and make it look good with CSS.
You only need to make UI using JS only for complex components.
Case study: I tried porting an old strategy game from pygame to web using codex. Codex decided that it doesn't need any library and raw-dogged HTML. It was able to match look and feel of an old UI with a very simple, maintenable code. Doing that with just CSS sounds kinda tedious to me, but it can be done, and honestly it looks more maintenable than any UI library.
The problem is really folks trying to one-size-fits-all web projects. React basically exists to give a “good enough” GUI toolkit functionality in the browser. On the other hand, the web was built to publish content as in articles and information. Everything kind of sits somewhere on the continuum between static content and the interactive app experience. At the extreme it can be interactive “art” piece.
It is important to consider digital media (as in art medium) best to convey an idea. And folks should start with HTML/CSS. Minimally, JS can be used for client-side validation. But that’s not necessary since the server needs to validate anyway.
I think main reason people choose React now is "nobody was fired for choosing React". (That used to be a saying about IBM as a vendor.) People can build their career around React.
And it is better "job security" than pure HTML+JS: React apps need constant maintenance
For me, what the browser needs it's a good DX to build components, I kind of like Svelte and Vue in that regard, where your CSS, HTML and JS lives all together.
Writing a CSS file with the time will start growing and at some point will be very hard to know what to keep and what to delete, I think this is the reason why tailwind became popular, because it keep only the used rules, now the downside is that now your html elements has ton of classes.
Re-creating minimal framework API is not that difficult. What is difficult is what frameworks are doing under the hood. To support ui=f(state) in the browser/dom you beed something more complex than pub/sub. The problems are 1. We need to calculate what part if UI depends on changed state 2.apply this to a dom efficiently. What author describes is a simple push-only reactivity which does not scale well because all updates are propagated to all subscribers. Libraries Solid and Vue (very simplified) are doing push-pull (push-pull-push) reactivity which involves traversing the dependencies for peace of the state, marking it as dirty and then recomputing the value of that peace of state. Another issue is DOM performance. When changing many parts of the state we want to apply changes to the DOM nodes in predictable and uniform fashion, avoid many writes/repaints, which usually involves some kind of scheduler. Another problem on top of that is measuring DOM before or after changes were applied (getBoundingClientRect can force synchronous layout/reflow, ResiseObserver is a more modern approach) and reacting to that.
Having said that I’m all for using solutions appropriate for the problem, but what many of such tutorials miss is a better problem statement. Yes frameworks are heavy, yes we may not need them. But problems with doing complex DOM manipulations are still there.
- List of comments/recommendations/pictures (sometimes in a tree), virtualized and lazy loaded
- A video
- A comment box
The irony is that this is super easy to represent in HTML semantically (HTML was literally made for this).
The 'virtualized and lazy loaded' part should've been a HTML standard (or more generally, partial page updates, like when you submit a comment) and then we'd have basically very little reason to do Javascript at all.
The irony of React is that what it does is make a mess of HTML, and allows you to ship your own semantic model in JSON/JS which will then get unpacked on the client into some display HTML.
This is accentuated by the fact that React's support for virtualization is really quite poor, as it assumes that you have the 'state' in RAM, and you have to go out of way to use third party libs that can handle both partial state and partial display.
Edit:
Apparently I'm not the only one who thinks this, and Chrome/W3C seems to experimenting with something similar:
> The only potential issue with immediate mode is performance, since we need to re-render the entire UI on every state change.
Notice that in the task list example, if you leave some text in the input and check off a task, your text will be erased.
If we could truly treat the DOM as an immediate-mode UI, there would be no need for React/Preact/ Svelte/Solid/etc. But it turns out there are a fair number of quirks like this that preclude such simple replacements from working correctly.
(That said: I am all for experiments like this for learning, or for fun, or to try out unexplored framework design space!)
I ran a web dev company for decades. There are two web sites we created I would bet money you have visited so we weren't a small, one-off shop. We had never seen the need or desire to use React. We couldn't understand why anyone else would use it either. It was too big and too complicated versus just using the fundamental elements of programming for the web. So there.
Counterpoint: I'm currently employed building many apps for an enterprise you'll never visit. React+Typescript rewrite of our legacy "lean" Javascript has measurably improved the productivity of the team and the maintainability of code.
I can't imagine going back to learning bespoke libraries written by (especially!) hotshot cowboy coders. React makes everybody on the team better, and literally no customer has ever complained about the extra 200ms load time.
A major feature of React (like Java) is the ability to reduce the blast radius of errant colleagues through much stricter boundaries.
In the AI/vibe coding world this becomes even more valuable, on top of the fact LLMs are well versed in React already and don't need to eat context to understand it.
The problems are people then assume that because you're using React you must use next.js, vite etc. Then you're in trouble.
I think the author is saying a potential issue of a vanilla implementation of immediate mode. The following text explains that there is already many optimizations for this
> I think the author is saying a potential issue of a vanilla implementation of immediate mode.
And you (pedro_movai) == the author? So you are saying that potential issue? Sorry, technically you are right that the author is saying that but I got confused with you referring to yourself in third person. Nothing wrong with it. Just want to confirm if you're the author.
That's called an "abomination". The advantage of immediate mode is not storing a DOM. If you write immediate-mode code but it stores a DOM anyway then it's the worst of both worlds.
Isn't that literally React? I find it so immeasurably funny that in his effort to simplify code and invent immediate mode UI, Casey Muratori accidentally invented React
I think the high quality of the author's website is more relevant than theoretical concerns about performance.
Loading performance on the s website is quite good, and it appears to be using the JS library described in the blog post. I also saw that pages were loading around 60kb or JavaScript or less.
Also, from looking at the author's GitHub profile, it appears that they do not make significant use of LLMs and actually took the effort to write the code by themselves.
Regarding small pages - there's a certain size threshold where the entire HTTP response is sent back in the first TCP window for most client OSes, which allows minimum round-trips.
React isn't immediate mode. The dependency array, amongst other things determine what redraws and what doesn't. Yes, you can make React look like it's running in immediate mode, fully redrawing on every frame. But if you do that then you're probably in the wrong line of work. Maybe try management.
This post is impressively bad. It's been AI generated and the AI doesn't even know what immediate mode is. Did you use Grok?
I wish that before someone writes yet another article on "You don't need React" they can do a Google search to look up all the 1000 articles already written with the same arguments and the discussions around that.
unlike many of the commenters, I found this pretty nice and educational, there are full stack developers who cannot write anything outside React and do not even understand why React was created in the first place
I agree that React gets overused for many websites, but I'm not sure about this project over something like Astro.
I built an Astro Starter that uses just CSS in a scalable way for simple websites. Content is written in Markdown / MDX files. Design tokens are set in var.css. Other stuff is defined in one config.ts file.
Global styling is set though a few CSS files (Global, Reset, Var, Util, Markdown), components are scoped in styling but utilise the tokens.
A 50-100 LOC library that provides 80% of what React does is very easy to create.
I created one of those 14 years ago before React, to replace things like mustache and similar templating languages. That doesn't take away from your effort. It's a good idea for anyone to recreate their favourite lib to understand the principles.
Posts like this remind me how much JavaScript could benefit from something like Lisp's quasiquote. JSX is a terrible creation but templating in JS is pretty unpleasant without quotation operators. That said, I think Mithril's 'm' function did it the best of all libraries I have seen.
And that’s fine. It’s easy to replatform later with LLMs.
With vanilla you typically do class based components for organization (which can be quite clean in type=“module” with import/export) and you end up writing a complex View type class that does all the DOM manipulation.
You gotta hide the framework somewhere or you’ll repeat yourself a lot
Bonus points at scale you invent a templating solution. Maybe you use JSX
on the backend this obviously runs only once. the main issue with backend frameworks is, that the request and other context data has to be passed through the business logic + VIEWS. leads to tight coupling of all the components as it is downstream from the data structure + presentation
this is resolved by resolving the html template strings at the end and allowing functions (that also produce html template snippets) to be passed into the template strings.
If you are going to advocate against something, the alternative you propose needs to be better in some important area, other than "not that thing".
What is the thing you hate about react, and what is the thing you require in an alternative?
I hate bloat and require lightness, so I use Preact.
I also hate complexity and difficult to read stuff, so when I look at the proposal here, I don't see anything appealing, other than "look! it's not react!"
I do understand that's what they have in mind. But that's not what immediate mode means. Function application is not the only criteria.
We are talking about rendering into a browser by twiddling DOM APIs. That's definitely retained mode rendering and repeatedly claiming it's immediate mode doesn't make it true
If you compare the code using the presented framework to code using say immediate mode library like ImGui[1][2] for example, the presented framework does indeed seem to be exposing a retained mode API.
But I can’t shake the feeling that discussing this stuff is indeed void on the AI era. I considered my self a JS/TS/React expert dev, derivate, state duplication, url as a source of truth…
All of these points feel void now, I have not checked/cared how my React code looks like in months now. I only care about more high level stuff like schema definition and overall wiring.
You're showing your ignorance here. Different frameworks have different ways of working that can lead to better or worse performance on client devices, no matter what Fable spits out for you. You owe it to your users to care.
That useState hook doesn't work the way you think it does. It's going to reset state every time the hook is invoked on rerender. It also isn't triggering a render, which is the entire point of the hook.
I think React is way overused but every time you see a blog post with "replace it with this pet mini project instead" I groan because you know the reaction is going to be "what about X feature", and of course the mini framework doesn't do it.
I've found the perfect balance to be Astro with Preact. You will inevitably have some piece of functionality that's complex (say, a contact form) and you can lean into Preact for that. But for the vast majority of a content-heavy site you can just use Astro and skip the client-side bulk entirely.
STANDARD DISCLAIMER ANY TIME I COMMENT ON FRONT END DEV: your project may be different. A blog or a shopping site and have extremely different requirements to a full Gmail-style web app. There does not need to be a one size fits all answer.
I kind of agree but also
> what about X feature
To which I argue unequivocally YAGNI
95% of apps people actually develop are really just CRUD and could easily get by without JavaScript on the front end at all. They certainly don't justify the layers of complicated state maintenance React and similar systems entail.
> 95% of apps people actually develop are really just CRUD and could easily get by without JavaScript on the front end at all.
Nearly every app has some version of CRUD, but you're conflating the persistence model with the interaction layer. Just because an app is CRUD doesn't mean it doesn't require complex, client-side interactions.
I'm building an EMR and there is a large portion that is CRUD, but that doesn't mean I don't rely on enormous forms or don't need client-side validation.
And not only, but "you could get by" is true of a lot of things. I could get by with full-page refreshes every time a form is saved or a chart is pulled, but that doesn't mean it's the best option or that the users won't notice it.
This argument has big "I seldom build web apps but have a lot of opinions on them" energy. Like, the idea that I, myself, have seen enough of the different projects and use cases for the web that I can unequivocally state something like "95% of apps don't need a front-end framework" reeks of ego. You just flat out haven't. The field is enormous.
95% of the frontend apps can easily be made with AI. Doesn't matter in which framework.
It matters when you use an headless CMS from the MACH architecture enterprise culture, and they only support a specific SDK, which is mostly Next.js/React.
- https://macharchitecture.com/
- https://www.sanity.io/studio (a possible example)
Something being 'made with AI' in no way translates to the framework used becoming irrelevant.
> You will inevitably have some piece of functionality that's complex (say, a contact form)
And this is why it pays to just learn React. It's never been easier. A contact form is not that complex. React won. The LLM can author it pretty well at this point, probably better (and cheaper) than at least 50% of FE devs. Treating a "contact form" as a potential fork in the road in 2026 is just sad.
> And this is why it pays to just learn React.
It pays you, the developer, yes. It does not benefit the user.
> A contact form is not that complex.
Correct, my example was not a great one, just the first one that came to mind. Using a giant JS framework to do something as simple as a contact form should be seen as a failure of development culture.
> Treating a "contact form" as a potential fork in the road in 2026 is just sad.
Choosing React as a one size fits all because it suits you as the developer, with little to no consideration about what suits the user... that's what is sad in 2026. But it's where web development culture is. There should be no "React won". There is no need for there to be one winner. As developers we ought to be able to tailor solutions instead of be lazy.
SSG is solved for in React. The user is going to cope with a few extra KB of JavaScript just fine. Just like they cope when they watch Youtube or stream or w/e all day long. It's 2026. Data is cheap. The user really doesn't care until they hit your contact form that sucks because you tried to reinvent the wheel instead of just using React. Or when you need to branch out should your site ever do something actually interesting and this problem then gets even worse.
It's not hard anymore to get over the React learning curve. For a content heavy static site just use a good enough LLM and the right starter template and you'll never have to code, SSG and bundle size will just work, lighthouse will be all green. Your users do not care about the tech. The idea that React is the enemy of users sounds like a skill issue on your part. Good thing LLMs exist now, so you don't have to fret about the alleged perverse incentives. For a static site in 2026 a FE dev should be largely irrelevant. What could be more user friendly? Seems like big bad React has redemocratized the web right under your nose.
Data is cheap and fast, CPU time still isn't. I've profiled numerous web sites on slow Android devices and they absolutely crawl when parsing JS and giant JSON blobs required to hydrate a page that's one big React app for no real reason.
> It's not hard anymore to get over the React learning curve.
> The idea that React is the enemy of users sounds like a skill issue on your part.
You're contradicting yourself here. Is React incredibly simply to learn or is optimizing it for speed a "skill issue"? I'm not talking about myself here, I'm talking about the state of the industry. The web is full of slow-as-shit web sites that are held back by JS and JSON parsing. You don't need to convince me they aren't because I've seen it with my own eyes time and time again. As an industry it's become far more important to fart out basic functionality and move on than take any kind of pride in your output. It's sad and I'm not going to apologise for caring.
> You're contradicting yourself here.
>> Good thing LLMs exist now
It's not a contradiction. You were/are unskilled. But it doesn't matter anymore. Just wanted to help you understand this is not a contradiction. You sound like someone who made your mind up about "the state of the industry" a long time ago. While this replay of debates settled a decade ago has been a nice, I must wish you a good day now. I don't mean to be too harsh. Ultimately you can use whatever tool you like, obviously. If it works for you then it works for you. Often that is what matters most.
Thanks for the trip down memory lane.
> You were/are unskilled.
You're being ridiculous and trying to being personal insult into a conversation. I have, numerous times, asserted that I am discussing the state of the industry as a whole. I don't understand why you've repeatedly tried to misinterpret this, it's in extremely bad faith.
Every time I see this kind of pet project I see whether they are addressing any actual issue of React or whether they don’t like React merely because it is popular. I especially look at whether they do state management or VDOM diffing differently from React. I find that if a project had done that, it is likely something the author was very proud of and was front-and-center. And there have been some interesting ideas in pet projects like using languages with macros so the build tool does part of the diffing. This one replaced VDOM with… nothing at all?
> Every time I see this kind of pet project I see whether they are addressing any actual issue of React
Then I think you're looking at those projects from the wrong angle. The point the author is making is that React is often used in situations where VDOM or complex state management aren't actually necessary. If all you're doing is e.g. validating a few form fields then neither is needed. But people use it anyway because React has become the default for frontend engineering no matter what.
Same, at least for smaller projects where I want some kinds of interactivity, but also a good static rendered page first and just the ease of writing, Astro+preact has been very nice.
I switched my small personal projects to preact in a (temporarily successful effort) to get one under 14kb after re-reading [0] a year or two back. So far haven't run into any limitations.
If you need React, then go ahead and use React, but I'd encourage people to try a simple s/react/preact/g on their websites. (Okay, not quite as simple as that, but even if you don't like vibe coding, this is something an LLM can be trusted with.)
[0] https://endtimes.dev/why-your-website-should-be-under-14kb-i...
Why should a contact form mean complex functionality? You can do it in plain HTML, even - is that complex already? Because validation? Like that (Google, it's you) address checking unable to place my house in the right village? Like phone number validation forcing you to add a 0 where no 0 is required to call me? And don't start me on streets and all that, or middle names, or dates or or or. Validations of form inputs are almost always a stupid yak shaving exercise, yet here we are.
Actually browsers already come with a minimal UI library. It's called HTML + CSS + JS.
There's actually no need to make any wrappers - you describe what you want in HTML and make it look good with CSS.
You only need to make UI using JS only for complex components.
Case study: I tried porting an old strategy game from pygame to web using codex. Codex decided that it doesn't need any library and raw-dogged HTML. It was able to match look and feel of an old UI with a very simple, maintenable code. Doing that with just CSS sounds kinda tedious to me, but it can be done, and honestly it looks more maintenable than any UI library.
The problem is really folks trying to one-size-fits-all web projects. React basically exists to give a “good enough” GUI toolkit functionality in the browser. On the other hand, the web was built to publish content as in articles and information. Everything kind of sits somewhere on the continuum between static content and the interactive app experience. At the extreme it can be interactive “art” piece.
It is important to consider digital media (as in art medium) best to convey an idea. And folks should start with HTML/CSS. Minimally, JS can be used for client-side validation. But that’s not necessary since the server needs to validate anyway.
I think main reason people choose React now is "nobody was fired for choosing React". (That used to be a saying about IBM as a vendor.) People can build their career around React.
And it is better "job security" than pure HTML+JS: React apps need constant maintenance
Another reason are SaaS and iPaaS SDKs having it as favoured stack, which kind of builds on your assertion.
For me, what the browser needs it's a good DX to build components, I kind of like Svelte and Vue in that regard, where your CSS, HTML and JS lives all together.
Writing a CSS file with the time will start growing and at some point will be very hard to know what to keep and what to delete, I think this is the reason why tailwind became popular, because it keep only the used rules, now the downside is that now your html elements has ton of classes.
Re-creating minimal framework API is not that difficult. What is difficult is what frameworks are doing under the hood. To support ui=f(state) in the browser/dom you beed something more complex than pub/sub. The problems are 1. We need to calculate what part if UI depends on changed state 2.apply this to a dom efficiently. What author describes is a simple push-only reactivity which does not scale well because all updates are propagated to all subscribers. Libraries Solid and Vue (very simplified) are doing push-pull (push-pull-push) reactivity which involves traversing the dependencies for peace of the state, marking it as dirty and then recomputing the value of that peace of state. Another issue is DOM performance. When changing many parts of the state we want to apply changes to the DOM nodes in predictable and uniform fashion, avoid many writes/repaints, which usually involves some kind of scheduler. Another problem on top of that is measuring DOM before or after changes were applied (getBoundingClientRect can force synchronous layout/reflow, ResiseObserver is a more modern approach) and reacting to that.
Having said that I’m all for using solutions appropriate for the problem, but what many of such tutorials miss is a better problem statement. Yes frameworks are heavy, yes we may not need them. But problems with doing complex DOM manipulations are still there.
The whole idea of SPAs is just unfortunate most of the time. If you look at the list of the most popular websites:
https://en.wikipedia.org/wiki/List_of_most-visited_websites
Basically all of them can be decomposed to:
- List of comments/recommendations/pictures (sometimes in a tree), virtualized and lazy loaded
- A video
- A comment box
The irony is that this is super easy to represent in HTML semantically (HTML was literally made for this).
The 'virtualized and lazy loaded' part should've been a HTML standard (or more generally, partial page updates, like when you submit a comment) and then we'd have basically very little reason to do Javascript at all.
The irony of React is that what it does is make a mess of HTML, and allows you to ship your own semantic model in JSON/JS which will then get unpacked on the client into some display HTML.
This is accentuated by the fact that React's support for virtualization is really quite poor, as it assumes that you have the 'state' in RAM, and you have to go out of way to use third party libs that can handle both partial state and partial display.
Edit:
Apparently I'm not the only one who thinks this, and Chrome/W3C seems to experimenting with something similar:
https://developer.chrome.com/blog/declarative-partial-update...
But this should've been a W3C standard in like 1999.
> The only potential issue with immediate mode is performance, since we need to re-render the entire UI on every state change.
Notice that in the task list example, if you leave some text in the input and check off a task, your text will be erased.
If we could truly treat the DOM as an immediate-mode UI, there would be no need for React/Preact/ Svelte/Solid/etc. But it turns out there are a fair number of quirks like this that preclude such simple replacements from working correctly.
(That said: I am all for experiments like this for learning, or for fun, or to try out unexplored framework design space!)
I ran a web dev company for decades. There are two web sites we created I would bet money you have visited so we weren't a small, one-off shop. We had never seen the need or desire to use React. We couldn't understand why anyone else would use it either. It was too big and too complicated versus just using the fundamental elements of programming for the web. So there.
Counterpoint: I'm currently employed building many apps for an enterprise you'll never visit. React+Typescript rewrite of our legacy "lean" Javascript has measurably improved the productivity of the team and the maintainability of code.
I can't imagine going back to learning bespoke libraries written by (especially!) hotshot cowboy coders. React makes everybody on the team better, and literally no customer has ever complained about the extra 200ms load time.
A major feature of React (like Java) is the ability to reduce the blast radius of errant colleagues through much stricter boundaries.
In the AI/vibe coding world this becomes even more valuable, on top of the fact LLMs are well versed in React already and don't need to eat context to understand it.
The problems are people then assume that because you're using React you must use next.js, vite etc. Then you're in trouble.
> The only potential issue with immediate mode is performance, since we need to re-render the entire UI on every state change.
How this made it to HN front page?
I think the author is saying a potential issue of a vanilla implementation of immediate mode. The following text explains that there is already many optimizations for this
> I think the author is saying a potential issue of a vanilla implementation of immediate mode.
And you (pedro_movai) == the author? So you are saying that potential issue? Sorry, technically you are right that the author is saying that but I got confused with you referring to yourself in third person. Nothing wrong with it. Just want to confirm if you're the author.
Nothing in HTML is immediate mode. Any framework with a DOM is retained mode, by definition.
But the api is trying to simulate immediate mode using DOM
That's called an "abomination". The advantage of immediate mode is not storing a DOM. If you write immediate-mode code but it stores a DOM anyway then it's the worst of both worlds.
Isn't that literally React? I find it so immeasurably funny that in his effort to simplify code and invent immediate mode UI, Casey Muratori accidentally invented React
How would you do layouting without a tree structure in place? Not really sure what you're getting at here
You have discovered one advantage of retained-mode designs.
"I think the author is saying" - Wait, are you not the author? This seems dishonest.
I think the high quality of the author's website is more relevant than theoretical concerns about performance.
Loading performance on the s website is quite good, and it appears to be using the JS library described in the blog post. I also saw that pages were loading around 60kb or JavaScript or less.
Also, from looking at the author's GitHub profile, it appears that they do not make significant use of LLMs and actually took the effort to write the code by themselves.
Even this is more complex than many websites need to be. A few months ago I wrote about why I often create pages in pure html and css.
https://joeldare.com/why-im-writing-pure-html-and-css-in-202...
Regarding small pages - there's a certain size threshold where the entire HTTP response is sent back in the first TCP window for most client OSes, which allows minimum round-trips.
Yes. You might be talking about the 14Kb problem.
https://joeldare.com/the-14kb-problem
That post is so abbreviated as to be wrong in a few places (for instance, it's not that your "HTML" needs to be under 14kB, the packet, incl. headers needs to be). Fuller link: https://www.tunetheweb.com/blog/critical-resources-and-the-f...
It's not worth chasing except for personal curiosity and some truly extreme cases.
The response needs to be. The response can span several packets, up to the maximum receive window chosen by the client's OS.
React isn't immediate mode. The dependency array, amongst other things determine what redraws and what doesn't. Yes, you can make React look like it's running in immediate mode, fully redrawing on every frame. But if you do that then you're probably in the wrong line of work. Maybe try management.
This post is impressively bad. It's been AI generated and the AI doesn't even know what immediate mode is. Did you use Grok?
I wish that before someone writes yet another article on "You don't need React" they can do a Google search to look up all the 1000 articles already written with the same arguments and the discussions around that.
unlike many of the commenters, I found this pretty nice and educational, there are full stack developers who cannot write anything outside React and do not even understand why React was created in the first place
Nice! positive comment
I agree that React gets overused for many websites, but I'm not sure about this project over something like Astro.
I built an Astro Starter that uses just CSS in a scalable way for simple websites. Content is written in Markdown / MDX files. Design tokens are set in var.css. Other stuff is defined in one config.ts file.
Global styling is set though a few CSS files (Global, Reset, Var, Util, Markdown), components are scoped in styling but utilise the tokens.
Link: https://starter.bryanhogan.com/
GitHub repository: https://starter.bryanhogan.com/
A 50-100 LOC library that provides 80% of what React does is very easy to create. I created one of those 14 years ago before React, to replace things like mustache and similar templating languages. That doesn't take away from your effort. It's a good idea for anyone to recreate their favourite lib to understand the principles.
Posts like this remind me how much JavaScript could benefit from something like Lisp's quasiquote. JSX is a terrible creation but templating in JS is pretty unpleasant without quotation operators. That said, I think Mithril's 'm' function did it the best of all libraries I have seen.
You don’t need React until you do.
And that’s fine. It’s easy to replatform later with LLMs.
With vanilla you typically do class based components for organization (which can be quite clean in type=“module” with import/export) and you end up writing a complex View type class that does all the DOM manipulation.
You gotta hide the framework somewhere or you’ll repeat yourself a lot
Bonus points at scale you invent a templating solution. Maybe you use JSX
I agree - check out VanJS for minimal reactive library: https://vanjs.org
https://geajs.com/ done.
Minimal UI library results in maximal App code, which makes it a good usecase for why you're better of with React/Vue
I looked at the author's website, and a typical page downloaded around 60kb or less of JavaScript.
Who knew the largest community in Web tech had it wrong all this time?
here is one that uses the fact that the raf loop will slow down when the tab is out of sync. makes state handling much easier. similar to dearimgui
frontend example https://github.com/spirobel/counter/blob/master/frontend/cou...
on the backend this obviously runs only once. the main issue with backend frameworks is, that the request and other context data has to be passed through the business logic + VIEWS. leads to tight coupling of all the components as it is downstream from the data structure + presentation
this is resolved by resolving the html template strings at the end and allowing functions (that also produce html template snippets) to be passed into the template strings.
this decouples views, business logic and data
https://github.com/spirobel/mininext/blob/master/docs/archit...
worked on this for the last few years. But it is still only a few hundred lines of code that you can understand in a few hours
honestly if you go vanilla just go vanilla, don't imitate the vdom but worse
use query selectors and manually do fine grained updates
i don't think it works for everything, but a few years back i started doing it for side projects and never really missed any of the frameworks
If you are going to advocate against something, the alternative you propose needs to be better in some important area, other than "not that thing".
What is the thing you hate about react, and what is the thing you require in an alternative?
I hate bloat and require lightness, so I use Preact.
I also hate complexity and difficult to read stuff, so when I look at the proposal here, I don't see anything appealing, other than "look! it's not react!"
The blog post page, which appears to use the author's library only loaded 58.6 kB of JavaScript, which is quite low.
Preact is 3kb. The size argument doesn’t hold weight.
React is popular because it's pushed by Meta.
React is popular because it mainstreamed components as a concept to modularize and reuse UI.
Maybe it’s not needed. But why not just use React anyway?
> immediate mode
You keep using that word. I do not think it means what you think it means.
React most definitely falls under what would be called a "retained mode" of some sort. And the UI library described here is also.
I think the author is referring to immediate mode because of
UI=f(state)
I do understand that's what they have in mind. But that's not what immediate mode means. Function application is not the only criteria.
We are talking about rendering into a browser by twiddling DOM APIs. That's definitely retained mode rendering and repeatedly claiming it's immediate mode doesn't make it true
If you compare the code using the presented framework to code using say immediate mode library like ImGui[1][2] for example, the presented framework does indeed seem to be exposing a retained mode API.
[1]: https://pthom.github.io/imgui_explorer/
[2]: https://github.com/ocornut/imgui
With that argument, every UI is retained mode because everything gets saved in the GPU's scan out buffer.
Isn't React specifically an attempt to build an immediate mode UI on top of the DOM?
I read the blog post, and your framework looks great.
I think React is overused, and I'm happy to see people support alternatives.
After looking at the code in the tic tac toe example I am now 100% convinced that I need React.
React and systemd feel the same way to me: both really sensible ideas at the core, surrounded in layers of bullshit.
What is the point of this?
Discussing language preferences at all sounds very pre-Fable era
Tell me, mister Bofadeez Nutz, why does every non-AI post have to have these dumb "hurr durr just use AI" comments?
I “agree” with your “sentiment”.
But I can’t shake the feeling that discussing this stuff is indeed void on the AI era. I considered my self a JS/TS/React expert dev, derivate, state duplication, url as a source of truth…
All of these points feel void now, I have not checked/cared how my React code looks like in months now. I only care about more high level stuff like schema definition and overall wiring.
You sound like me before I tried Fable.
You sound like most people I know that only have a Claude subscription. GPT-5.6 is so much better bang for the buck.
You sound like me after I tried Cocaine.
How many times have you abused cocaine?
How many times have you abused fable?
Framework, not language. And framework choice still matters.
So do programming languages, fwiw. The argument of "we have Fable, X doesn't matter anymore" is a severe case of delusion.
I would have agreed before I used Fable. It's not even vibe coding anymore. It's just coding.
E.g. Just always use Rust instead of Python.
Tomato tomato. English is my favorite new programming language.
You're showing your ignorance here. Different frameworks have different ways of working that can lead to better or worse performance on client devices, no matter what Fable spits out for you. You owe it to your users to care.
Have you tired using several billion Fable tokens in production? It will do whatever you want.
So you can still have your opinions on what works best. But it doesn't matter much, it's not 2025 anymore.
E.g. always use Rust instead of Python
Ah. I see. A troll account.
That explains the 7.5 year account age and only 2 karma.