FRESH
Hacker News
Hacker News Hacker Newshttps://htmx.org/essays/when-to-use-hypermedia/
Also, please try unpoly:
It’s another excellent hypermedia oriented library
Edit: the article is actually not nearly as unreasonable as I thought based on the just-f*king-use template. Still prefer a chill vibe for htmx though.
Of course it will work. I can vibe code the most terrible web framework you have seen within 20 minutes and claim it is better than React, but what does it prove?
> You write zero JavaScript > The whole library is ~14kb gzipped
Oh sure, if there is nothing else in your UI, and if your Python/Go/whatever backend server has 0 dependency, which almost never happens.
My startup did.
And now we’re going to rip it all out and move to a React front-end.
HTMX makes response handling much more complex. Every endpoint returns 3–5 different HTML fragments. Frontend and backend must agree on every scenario — success, validation errors, system errors, partial updates, full reloads.
And HTMX is still a fairly obscure library. The documentation and examples are lacking, there isn’t a real set of established best practices at scale, and not for nothing, LLMs aren’t great at it.
React is mature, used at scale, provides separation of concerns, and is great for agentic AI coding. HTMX has its place for simple projects, but for anything non-trivial, it’s a no for me.
No bundler required, no compilation step.
And now I'm at the breaking point. So I'm planning to move to tailwind and Go templates, but honestly, i was hopeful for htmx, so I need to properly see the usecase. Which i don't know is this. It reminds me of Angular a lot...
If your solution is actually good, it will get adopted eventually...
Forget React, there's still stuff written in jQuery and JSP.
Why the rush to convert everything - you're not a missionary on a mission, just build your stuff in stuff you like?
The attack on npm is ridiculous, when (apart from introducing a permanent vulnerability in the form of a runtime dependency on a third party site), you still need npm for htmx.
The idea that HTMX removes all the complexity is false. However it does remove some of it, and moves the rest onto the backend.
I find the backend easier to work with, so that's a win for me.
And a batteries included framework (like Laravel or Django) makes HTMX even more appealing. They already have everything you need! :)
In my experience, a lot of SPAs transfer more data than the front-end actually needs. One team I worked on was sending 4MB over the wire to render 14kb of actual HTML. (No, there wasn't some processing happening on the front-end that needed the extra data.) And that was using graphql, some dev just plunked all the fields in the graphql query instead of the ones actually needed. I've seen that pattern a lot, although in some cases it's been to my benefit, like finding more details on a tracking website than the UI presented.
However, I’ve recently made the difficult decision to rewrite the frontend in React (specifically React/TS, TanStack Query, Orval, and Shadcn). In a perfect world, I'd rewrite the python backend in go, but I have to table that idea for now.
The reason? The "LLM tax." While HTMX is a joy for manual development, my experience the last year is that LLMs struggle with the "glue" required for complex UI items in HTMX/Alpine. Conversely, the training data for React is so massive and the patterns so standardized that the AI productivity gains are impossible to ignore.
Recently, I used Go/React for a microservice that actually has turned into similarly complex scale as the python/htxm app I focused on most of the year, and it was so much more productive than python/htmx. In a month of work I got done what took me about 4-5 months in python/htmx. I assume because the typing with go and also LLM could generate perfectly typed hooks from my OpenAPI spec via Orval and build out Shadcn components without hallucinating.
I still love the HTMX philosophy for its simplicity, but in 2024/2025, I’ve found that I’m more productive choosing the stack that the AI "understands" best. For new projects, Go/React will now my default. If I have to write something myself again (God, I hope not) I may use htmx.
Any big reason to use HTMX instead? Is Turbo not really discussed much because of it's association to RoR?
I get server-side rendering. I can boot my server, and everything is there. If my model changes, I can update the view. It's cohesive.
I get client-side rendering. The backend returns data, the frontend decides what to do with it. It's a clear separation. The data is just data, my mobile app can consume the same "user" endpoint.
This seems like a worst-of-both-worlds paradigm, where the backend needs to be intimately aware of the frontend context. Am I not getting it or is there a massive implicit coupling?
Now if I need to display the same "user" data twice, in different formats, on my website. Say, as a table in "My account", and as a dropdown in the menu bar. Do I need to write two endpoints in the backend, returning the same data in two different formats?
I really liked HTMX, and I thank the authors for this marvelous library!
I switched from Turbo to HTMX because the latter is much more flexible, and I try to avoid Node.js as much as possible, only using it to compile some JavaScript code for Stimulus.
I finally moved from HTMX to Unpoly for the following reasons:
1. Layer support: Unpoly makes it easy to create layers and modal overlays, saving a lot of time and JavaScript code. You can achieve the same functionality with HTMX, but you have to write more code.
2. JavaScript code is better organized thanks to up.compile hooks.
3. HTMX and Unpoly treat fragments slightly differently. With HTMX, you have to use an out-of-band feature to update multiple fragments together. With Unpoly, you can easily add them to the response (and declare them in the front end, of course).
In my opinion, Unpoly has a better-organized approach to everything. On the other hand, apart from the official documentation, it is difficult to find examples for some edge-case features.
Why not "just use HTML"?
First - simple use cases sure great. But imagine you have to update some element out of the from tree. Now you need to have OOB swaps and your HTML must contain that fragment.
Not just that your server template code now has to determine if it is HTMX request and only render OOB fragments if so.
Even at decent size app, soon it turns super brittle.
Yet to talk about complicated interfaces. Let's not go complicated just think of variants in an E-commerce admin panel.
3 variants with 5 values each these are 125 SKU rows that must be collapsed group wise.
htmx can do it but it's going to be very very difficult and brittle.
So it is surely very useful but it is NOT the only tool for all use cases.
The issue with htmx is that it is fairly prescriptive of how one should go about building dynamic interactions, and it becomes complex quickly if the dynamic interaction is more than trivial. I don't disagree with its philosophy at all (as I say, I use it for my own site) but it becomes an issue when my product owner tells me that I need to do some funny dynamic thing because it will make the business or clients happy (for some reason), and then it becomes a mission to wrangle it with htmx attributes. And I have to follow that, because as much as it pains me to say it, making stuff pretty and dynamic on the UI is an easy way to score points. It is one of those areas of enterprise software development which seems like a huge upgrade to non-technical people whilst not requiring too much effort.
The one thing raw JavaScript is quite well suited for is hacking together some DOM manipulation. I dislike JavaScript in every other domain except this - its in this arena where its leniency is very useful.
Given sufficient time and money (20+ years, and billions (trillions?) of dollars - which is what we've thrown at web apps) you could build GUI apps using the IRC protocol, but it will never work well.
LLM generated code probably tips the scales toward using react though. You can have the bots churn through all that boiler plate, it won't be any worse than what human react devs write, and keep the bots away from your mission critical code because it isn't all munged together like in a SSR app.
The only reason react seems to look beautiful is because it's compared to custom JavaScript state management rather than the server-state paradigm that probably makes most sense for over 90% of apps.
The only real place it makes sense is when you're Google or Facebook and those 15 milliseconds actually translate to quantifiable lost user attention.
If your app is actually useful and not just an attention farm, those 15ms are almost never worth your engineering team's sanity.
Well you didn’t write standard HTML attributes, you wrote custom attributes that are picked up by a JS framework, so potentially the worst of both worlds depending on your problem space.
Having tried HTMX a few times, the problem is firmly in creating a backend that feeds it properly. It’s a disjointed experience for anything more complicated than updating content.
It made the site look like a SPA without having to write any extra code. How cool is that.
HTMX resembled me this library. But it seems very narrow cased, there are only so many attributes and their values and you cannot implement anything else. While pjax is generic: you can attach it to any site which has links.
Also you cannot replace Vue (don't use React) with HTMX. For example, if you are making a diagram editor, HTMX won't be useful.
To be honest this might be a skill issue or something I haven’t understood properly with these frameworks.
Thats the thing I don't like. I don't want parts of the structure of my page coming from the backend server. I just want that to send data, as JSON and for the front end to handle that into whatever structure it deems suitable.
That way all of the front end code is in one place.
I personally prefer UIs with great encapsulation/composition, which used to be Vue, but with AI starting to write more of my UIs now I've switched to React/Next.js for new non-progressive UIs.
The main problem for me was storing/passing state between too many fragments. At some point some pages can become too complex to be manageable by HTMX, unfortunately. Lots of little fragments depending on each other, I began struggling to maintain a clear mental map of what was going on.
I'd say if React is more like functional programming, HTMX sometimes feels like GOTO programming.
I'm however more curious about going the other way, i.e. you start a project with Htmx which happily grows but after a while, a large feature is requested which inevitably changes the direction of the app into React use-case territory. I can't think of concrete example but you now have to work around it with htmx or commit to rewriting to React (or alternative). Wonder what are thoughts of people who had to deal with this
Hmmm.... I wonder why that is......
Too bad that the world insists on going nuts with JS everything.
Oh as a plus, AI agents are a lot more productive when dealing with server side logic.
HTMX serves fully baked HTML that needs to be created on the back-end (or front-end-facing servers)
That is more processing than sending the raw data to the front-end and baking the HTML there.
It is also more bandwidth (unless the JSON is more verbose than the HTML generated).
Lastly, I can generate different HTML fragments on the front-end from the same client-side state with the data only being transported once. How is that working out?
You can do this with plain old Javascript. Make a request, swap out the [inner | outer]HTML with the result. If you want a nice visual transition, wrap the swap in a startViewTransition(). Obviously, you need to be extra careful if you're using user-submitted HTML. Otherwise, it's fairly straight forward.
Just pure eval(). [1]
1. htmx.config.allowEval: defaults to true, can be used to disable htmx’s use of eval for certain features (e.g. trigger filters)
Agent one: handles the request and does tool calls
Agent two: reads the result and decides on quality vs a re-drive if it’s low quality
Agent three: decides how to present the information to the user, creates a collection of HTMX elements
HTMX hx-get is reliably, and beautifully rendering the result of the Agentic Workflow without any react, etc.
Very happy and passing quality gates. I love not having security alerts every week to patch because of some buried react dependency library
> And suddenly you've got:
> A package.json with 847 dependencies
> A build step that takes 45 seconds (if the CI gods are merciful)
> State management debates polluting your pull requests
> Junior devs losing their minds over why useEffect runs twice
> A bundle size that would make a 56k modem weep
No? React is surprisingly small, and if you're in dependency hell then fix that. The alternative is another idiom.
<button hx-post="/clicked" hx-swap="outerHTML">
No, I'll keep using hiccup and similar things that are just data and nothing more, no syntax, just functions and built-in data structures, then give me HTML as a string which consumers can do whatever with, and we're golden.
> the build time is over 30 seconds!
that's silly. 30 seconds building time is nothing compare to the accumulated time you wait for micro changes to your frontend.
for typical web development using react/vue/svelte you have hot code reloading, which can reload the current website < 1 seconds after you hit [Save] on your favorite editor.
for htmx to update, you have to wait for your whole server to reload (which can be way slower even you use interpreted languages like ruby or python, due to complexity of the framework you use).
not to mention it does not keep any state of the current website, make debugging way more troublesome compare to a js mature framework.
only people who never have to improve their website incrementally think htmx is a viable option. or insane people.
obviously, for some small websites with minimal interactions or no need to change the content very often, then htmx can be good, but for that case, vanilla js also works, and they do not need 14kb of excess baggage.
For things with heavy interaction (drag and drop, chat etc.), I find the code to make it work with HTMX is just too clumsy to work with as a mental model.
> The server just returns HTML (not JSON, actual HTML)
I like to separate presentation HTML from the data (returned from HTTP request). Some like to make backends that do nothing but serve the (singular) frontend, even running templates to make the HTML they return for easy consumption. That's not where I draw the line.
Dx resources must aim AI's attention having enormous technical documentation and be AI efficient in order to become mainstream.
I believe no other shiniest thing will ever make cognitive nest in humans. We are overloaded.
And every months a few of those modules try to exfiltrate your credentials…
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guid...
I switched to Hotwire/Stimulus and found it to be a significantly better implementation of the same core concepts. I'd highly recommend checking them out.
https://justfuckingusereact.com
With all the examples people keep using, I assumed it would be way smaller. 16kb minified is a lot.
Looking at the docs just now the core api seems reasonable, but it a lot larger than I'd assumed.
Install an open source admin panel and call it a day.
Oh now now, even senior devs do this too :)
You can use JavaScript for it.
I've been trying lately ruby templates with some occasional JS for this and it works great.
Or implement a straight forward web component.
They are standard, they work, they're great.
[1]: https://astro.build/
htmx is a great idea, but it's not necessary anymore. We're very close to invokers being baseline (and even that is just an extension of the "composedPath includes -> invoke" pattern), and that will take care of plenty of what htmx was designed to do. Between features like that, and web components, I'm very happy to stick with "plain HTML" (no frameworks; my web components do not draw from a core module or anything).
Also, just a suggestion because it's not technically wrong:
The example of sometimes needing an auto-completing search box is probably not the best one to use. I'm sure it's meant to say "you want the results to be queried from a server/database on every input", which you would certainly need javascript for. But the specific example of just having a search box autocomplete can actually be fulfilled with a datalist element. It won't dynamically re-query results, but it will filter based on input. So it's a muddy example, at best, and that's probably not great for the point trying to be made.
That'll be item #848 in my 847 line package.json.
And you can just return JSON from your API.
And you can add JSX later if you want.
CSS, HTML, Apache, MySQL and PHP.
Can someone explain something to me?
To my view, the single best idea React has is that it forces you to encapsulate the full set of states in your component - not anywhere else. For instance, if you have a "Comment" button that becomes "Submitted" after you click it, you must include that state inside the comment component. You can't hide it somewhere else (i.e., in the input field after you press cmd-enter). This is a huge thing for managing complexity - if you spread state updates to a number of different places, it becomes significantly harder to reason about how the button works.
Replacing the button with whatever the server responds with may sound simple, but it really makes it hard to answer the question of "what are all the states this button could be in". I mean, what if the server returns another button... which hits another API... etc?
The weird thing is that HTMX talks about Locality of Behavior (yay!), but doesn't seem to practice it at all?
BTW, one other thing:
> The ecosystem is why your node_modules folder is 2GB. The ecosystem is why there are 14 ways to style a component and they all have tradeoffs. The ecosystem is why "which state management library" is somehow still a debate.
> HTMX's ecosystem is: your server-side language of choice. That's it. That's the ecosystem.
Really? Python is my ecosystem. You know that people add stuff to their node_modules because they need it, right? It's not like we do it for fun. Where am I going to find a date-time picker in Python? Am I going to build it from scratch? Where is an autocomplete component in Python? Or mentions?
> Inspired by (and in joyful dialogue with) motherfuckingwebsite.com, justfuckingusehtml.com, bettermotherfuckingwebsite.com, and justfuckingusereact.com. Extremism in defense of developer experience is no vice! This site made by me. Does this all sound a bit like shallow slop? Yup, please help make it better.
I agree with you, and wrote a similar one for Markdown that you might enjoy. Same overall naming scheme. (Note: open the comments before you judge the use of a Web Component for rendering purposes.)
This part is said really well, and applies to anything. "It's just a single weekend" is a cure to the whole "wasting time" argument/excuse that I guess a decent amount of people have - at least I know I do.
Really well written article, having both when to use and when not to use, is a nicely balanced view.
I've built enough stuff in my time to know this is hyperbole at best and an outright lie at worst. I've never seen a team fail due to complexity. Team fails because the thing they built was wrong.
You should spend 99% of your time paranoid that the thing you're building is useful enough to justify its existence. Whatever tool you use along the way makes up the last 1%
You're not a fucking person, this is LLM output.
It starts with the overdone sweary thing then mentions it's overdone and says it's not gonna do it, and it's almost enough to make me think the article is going to offer someone's point of view. But once again the LLM has erased any point of view the author may have had going in (or prevented them from developing it) and replaced it with a mediocre infodump.
I think this is the 5th slop I've seen atop HN in 24 hours.
> This site made by me, with tongue firmly in cheek.
Well, the LLM ruined it, and you didn't even tell us it participated.
Why would you mock it on the client-side if HTMX makes it so simple??
> htmx is a library that allows you to access modern browser features directly from HTML, rather than using javascript.
From React website:
> React lets you build user interfaces out of individual pieces called components. Create your own React components like Thumbnail, LikeButton, and Video. Then combine them into entire screens, pages, and apps.
Apples and Oranges.
This means that in theory you (as a dev) don't need (to write any) js, nor do your users need to download a full page (for any interaction) like it's 1999. Your webserver replies with fully server-renderd HTML but just for the dom node (say a div) that you want to replace.
It's fun for very simple things, even great for extremely simple interactions modes. For interactive products, anything beyond simple CRUDs, it's madness.
Whenever you want to sprinkle a tiny bit of interactivity you'll have to choose between the path of least resistance (a small hack on HTMX) or a proper way. State management gets out of control real fast, it's the opposite of UI=f(state).
I've seen it go bad, then worse with alpine-js, and then completely ripped in favor of something where people can let the browser do browser things.
[edit for clarity]
> "But what about complex client-side state management?"
> You probably don't have complex client-side state. You have forms. You have lists. You have things that show up when you click other things. HTMX handles all of that.
On the other hand,
> I'm not a zealot. HTMX isn't for everything.
> Genuinely complex UI state (not "my form has validation" complex—actually complex)
But my interpretation is that any UI which displays the same data in just two places (like a "new notification" indicator as well as bolding new messages in an inbox, or a list of articles which can change dynamically as well as a count of the number of articles) is "complex" enough that you'll need client side state.
The article doesn't define any target audience in specific, so there you go.