> For better or worse, React is heavily overrepresented in the training weights, and “agent experience” is starting to matter more than developer experience.
Another winner takes all situation. This and the other post on agent tool use[1] makes me think we are in a period of great homogenization. I'm sure there's a group that's delighted about this but for me, it just makes me feel a little uneasy.
It's called alignment. And that's the correct word for it in my opinion.
The most important part is that we have to find a way to tackle the knowledge enshittification problem.
Because now that slopcode and slopposts are everywhere, the average quality level will decrease. We had the peak internet knowledge, and it's now gradually getting worse.
That's not a personal opinion, that's how attention matrixes and their keys/values computation works.
My wife's experience is managing teams and has been involved in creating corporate websites using contractors. She doesn't know anything about html at all. She just manages people.
She's now working on a biotech start-up pre-funding. They have pretty much zero money. They want to redesign their website.
I set her up with my deepseek key and she's been using flash-v4 with image detection (I think deepseek says it's still experimental). She just provides the current website, tells how she wants it to look and it works great. I mean it's obviously ai generated but she gets quick results and is able to tweek things to look better by just telling deepseek what to do. She was even able to change the text in one of their images to its new product name without having to contact the graphic designer.
This is a simple example. Basically it's one of those earlier jobs that you could get someone to bid on for $x but now you can do yourself for $20.
My mother has a psychology practice and wanted to update her 20 year old website.
I did the first few prompts (single static page, basically a brochure, don’t worry about hosting) into ChatGPT to get her started and a few days later she sent me a .zip which I uploaded.
Only took a while because she wasn’t paying for ChatGPT so had to wait for the free limit to reset quite a few times.
Not that I'm in favor of spending money unnecessarily, but I find it kind of surprising she didn't just spend the $20, given this was a website for her professional money-making business.
> I find it kind of surprising she didn't just spend the $20
If it because the poster's mother dislikes the idea of subscriptions, then I totally get it. (Though I admit I have no idea if this has anything to do with the poster's mother's decision.)
I'll go out of my way as much as I can to avoid signing up for anything that is a subscription because of how horrible the entire corporate world has abused subscription dark patterns over the past couple of decades.
These days just the packaging of anything into a subscription makes me assume there is intent to screw people out of money.
While signing up for LLM usage is fairly low risk as far as subscription lock-in goes, someone non-technical might not be aware of that fact.
> Presumably a nicer website could lead to more customers.
It depends what "nicer" means. Paying a web designer/developer to create a bespoke website doesn't guarantee that said website will have a better UI/UX and, most importantly, "convert" at a higher rate. Not every web designer/developer understands IA, content, performance optimization, etc.
The popular AI tools produce websites that are clearly AI-produced if you know the patterns, but that doesn't mean they can't be usable and quite effective at converting. And if you have a bit of skill with prompting, AI can be really good at working through performance bottlenecks, etc.
so. if website does not lead to customers - why do any of that? actually why have website? Once you say yes to website you imply having one is somehow valuable, so in that case why not spend 20 dollars.
Just because websites can exist. Sometimes things are done for fun, or just for the sake of it.
This perspective declares all the personal websites of people worthless, however, they are priceless.
Because people's curiosity and creativity is priceless as well.
Similarly, why create art and share it around if it doesn't bring in customers? Why make music, take photos, dance? Do we live to serve customers, or do customers' existence serve us to live a better life?
Of course, those are all great reasons to make a website. But this website was for a woman's psychology practice. I don't imagine she made it for fun or to express her creativity. The purpose of a psychology practice's website is to tell people about your psychology practice, so they will either sign up ($) or find information they need more easily (higher satisfaction → $).
If OP's mother runs a donation-funded pro-bono psychology practice then you can translate the dollar signs to units of happiness or whatever (which I would imagine is her real goal regardless).
This person is spending her time to create the website (with ChatGPT's help). How much is that time worth? Of course when it's your business hourly pay is tricky, but I imagine she values her time highly enough that $20 is not worth very much of it.
I know an accountant, who has a respectable amount of customers and enough income to live comfortably. Their company has a website which only provides a logo, address and contact information, in a 23 line HTML file, complete with all the white space.
If their company can do well with this website, her psychology practice can do as well.
Some of the best services I got was from professionals with no or mundane online presence. Instead, their abilities and expertise moved them to the top in their respective fields, and the whole community knows how and when to reach them.
So, maybe a shiny website is not a hard-requirement for a good psychology practice, IDK.
So, the thing is, maybe this was her way to explore new things and decompress while doing some mundane thing from her perspective for free. Sometimes it's not that $20 which you can pay without any negative side-effects, but the desire to see how far the thing goes without the token, or just because you don't feel any urgency to do the task, so that $20 looks like an unnecessary waste.
Yes, presumably that's why she did not update the website for 20 years!
If once she's decided to update the website—surely it's worth more then $20 at that point? Otherwise, she probably shouldn't be spending any time on it herself either, given the opportunity cost.
Obviously she decided otherwise, I just find it kind of surprising.
Like, in general, I'm always kind of surprised at what people are and aren't willing to pay for. People will suffer through horrible full screen ads in mobile games instead of paying $5 one time to turn them off. And they'll do this on a phone that costs $1,000+. I don't know their financial situation, but I would buy a cheaper phone and get rid of the ads...
From what I have seen, people's perspective on value lies on a very wide spectrum. My mom won't buy a YouTube premium subscription, because she said she likes the ads. It's a kind of discovery platform for her. It's not that I share the same perspective, but I respect her about that.
I know people who won't subscribe to a service, but go through great pains to get the thing the service offers (another brand, similar yet inferior service, etc.).
For example, if an application I use has ads (i.e. a game), and it needs a subscription to remove these ads, I'll simply won't install it, or get the paid version of it from the start. For me, playing a game with no end and paying monthly to "hide" advertisements doesn't worth it.
These things are very subtle and are sometimes about ethos, and not about value at all. For instance:
* You propose I buy your ¤4 overhyped toothbrush. If I spend 10 minutes digging around, I could find a sub-¤1 toothbrush. Effectively I'm doing 10 minutes of work at less than minimum wage in order to avoid owning the premium toothbrush. Do I do this? I might! The reason is that I want no part in your hype.
* You offer me a loyalty scheme that will genuinely save me money if I participate. I only have to pay attention to your offers and routinely identify myself and I might get 10% savings, all the time. I decline this because you don't get to use my brain cycles for your profit, even if I profit too.
* You offer me a free thing that happens to track me and funnels data to data brokers. I decline it even though it does me no actual harm at all, because I don't want you to make money from personal facts about me, even when the facts are robotically aggregated and anonymised. Ethics are at stake: ethically speaking, you can just piss off. That is worth more than a free gadget, and it's not even an equation.
Every static HTML pages built this way with ai is another WordPress site that wont get built by contractors just to be neglected, compromised, and used to launch phishing and ransomware attacks.
Still a win. Claude loves to write code from scratch instead of searching the web and pulling in dependencies, so you’ll likely end up with fewer garbage tier plugins written by people whose average skill level is put to shame by bootcamp grads; those people don’t know what a reverse shell is either.
Every static HTML pages built this way with ai is another WordPress site that wont get built by contractors just to be neglected, compromised, and used to launch phishing and ransomware attacks.
You mean the very web sites that the "AI" trained on?
If you think AI can't make buggy code, there are several dozen thousand comments on HN that disagree.
She wouldn’t be able to do any of that without you setting it up first. She would not know anything about deepseek, meaning of a folder with index.html file. Would not know even what to ask ChatGPT, “I need a website”?
You make it sound like easy task anyone can do, when you provided 99% of otherwise impossible for them to easily obtain key knowledge needed.
Any idiot can spend one minute now to acquire this amazing knowledge that you can just ask ChatGPT “I need a website”. If they don’t know now, sooner or later they’ll know from a short TikTok or something in the next couple of months/years. Not to mention AI companies are spending $$$ on ads to raise awareness; professionals already know they can create slides with LLMs, literally nothing’s different when it comes to most content websites. Yes, it’s an easy task any idiot can do. Head in sand, pretending it’s some sort of $10k for knowing where to hit situation isn’t going to reverse the trend.
You're not accounting for how deeply unpopular AI is with the general public. Would anyone be making those TikTok videos in the first place? Would OP's wife want to use AI if he wasn't there to advocate it?
Given that my nontechnical coworker has shared quite a few of these TikToks/Google Shorts with me, yes tons of people are making them. I don’t know what kind of alternative reality you live in, you can just go to any social media site and search for “create website with ChatGPT” etc. and check the view counts.
Plus the context here isn’t any random person, but people who intend to pay for website development for their businesses.
It's generally a little more complicated than that. Not up to writing paragraphs about it so I'll just leave this as a note so anybody that's interested can do their own research.
That was just part of her work. She was employee #11 at a now $4.5B biotech company so everyone that early on had to do everything. She worked for Novartis, for booz allen, and others. She usually hired McKinsey consultants for projects but the web stuff was for smaller companies. She managed a team in India for one of her companies that took care of the web stuff. Basically just look at any top-10 Pharma or government website and you might see what she managed.
i started a web development business in 1995 and have been coding web stuff for decades.
like the OP, all those years of learning how it works under the covers gives me a distinct advantage since i can easily tell what is slop and what is going to perform. as well as how to get AI to iterate and fix the things it isn't good at (yet). i feel like now i have super powers. i love it and can't get enough.
the fact that your wife can be productive at lower cost means that more products will be built and more opportunities will be made. instead of spending money on website design, her business can spend money developing biotech.
i welcome that, even if it means that things will change from the past and those that have made a career out of charging thousands will get replaced by a $20/month subscription.
Would you say this is a similar but perhaps bigger change to frontpage/Dreamweaver/web app frameworks? Those all drastically cut down tub development time for web pages and at the same time opened up new possibilities. I wonder how it'll pan out with AI
FrontPage, Dreamweaber and co seemed to fell into obscurity when HTML5 & CSS3 came up and from that point web development went much more complicated than that.
Granted, it seems it is easier now to give prompts to an LLM so it will spit things out than back then where you had to know the tool and fiddle with menus and thingies to sort of try to produce a result.
But the drawback seems to be the same for both - lots of unreliable and unoptimized code that will go live if you don't know what they're doing under the hood and can potentially cause issues in the future. Remember that all this tools do are sorts of interpolation. I've seen lately lots of "modern" websites with tons layers of !important in their CSS and who knows the monstrosities that they have with their JS.
Meanwhile I and all people like me still don't have a job because all of this and the alleged "thousands" I was getting for doing this things.
the thing you haven't come to terms with is that it is more important to ship than it is to worry about the issues in the future. i thought RoR taught everyone that.
the beauty of what's happening today is that AI might produce slop on the first try, but you can also refactor just as easily too. "clean up all !important and optimize the code for speed", is just as easy of a prompt and it'll go off and do a fantastic job at that. that is what the OP was talking about in his article.
you don't have a job because you're not adapting to this new world order quickly enough. i predict that there will be insane demand for porting existing applications off these over priced SASS contracts and building tailored products for businesses. as much as they are trying to convince people, you don't need a $5m/yr salesforce / servicenow contract when you understand the business model and can envision a better solution to the actual problem out of thin air.
i'm burning 500m+ tokens a day on a $200/mo subscription. it is creating 10000x more value for me. it is no longer how much you know HTML/CSS, it is how well you know how to get AI to do what you want it to do and the only way to develop that skill, is to do it.
> you don't need a $5m/yr salesforce / servicenow contract when you understand the business model
I'm sure many people will think that; and I'm also fairly confident that 5 years later they'll take their obscure, bug-ridden, vibecoded monstrosity that nobody really understands, throw it out, and get a salesforce contract. Technical debt doesn't disappear just because you have agents working for you; in fact, it is exploding right now, and I would bet good money that 2-3 years for now there will be a flourishing cottage industry of consultancies specialized in dealing with such monstrosities.
Context is important in a response like this. Giacomo is a Sr. Presales consultant at a SaaS company that sells "The AI operating system for modern Finance." Of course you're adverse to things because it is a direct threat to what your business offers today.
If I can vibe code up a solution that fixes my problems for $200, it doesn't matter what happens in 5 years. The rate of change that's happening today in the increase in quality of the output of the models, I can just tell it to fix whatever we have and make it better.
Spoken with the true arrogance of someone who has no idea about the real world of finance.
When your software has to produce legal documents, satisfy auditors, and certify a bunch of things to shareholders and governments, I'm sorry but your $200 vibecoded hack is not going to cut it. You will have to get it certified, to get it audited, to get it secured according to this and that legislation, to prove where it gets each and every number and how it modifies it, and then you will have to maintain it because standards and laws change every other day. Sure, get an agent to cook you up a full consolidation solution, see how quickly you get laughed out of the room by grown-ups.
Btw, I am not "adversed to things" - the stuff we do has agents built-in, that do all sorts of wonderful things. The difference is that we can demonstrate that they do them right, with financial intelligence standardized and battle-tested across thousands of customers all over the world. Our customers don't have to worry about rewriting their pile of vibed hacks, or about being humiliated by auditors, because we take care of all that stuff for them. That's worth more than the pennies you save by having cousin' Vinnie vibe up a multidimensional financial engine that takes 3 days to calculate trillions of useless intersections before falling over.
But more real... Giacomo fails to take into account the concept of when OpenAI decides to create an audited financial service that guarantees that their output is valid. ChatGPT Work is already a beginning of that future.
That's like saying Google offering GMail ended webmail - there are still plenty of webmail services around: some better, some worse, some more specialised. The same will be true of most sectors.
AI is another tool in the toolbox, not magic. It's not about being pro or against it, it's here already (hey, we sell it!) and obviously it will continue to get better; but it won't magically solve everything for $2.
That's great to know, Jon. I suspect non-US governments have a different opinion. Heck, even the US government likely has a different opinion. Not to mention Google competitors in any other sector.
I can see I'm under your skin as you're trying to make this personal. I have a long history of being early and correct with many things. What is transforming under your nose, faster than anything ever before, will take you for surprise, unless make a conscious decsion to wake up. Do yourself a favor and listen to what he's saying... https://youtu.be/NYFGCESmikA?t=2883
I was not the one who brought personal qualifications into a public argument, Jon. I guess you don't like it when your lack of specialized knowledge becomes evident.
As I said, I already know what AI is doing - I already use it just fine, I even help building and selling it. The impact it is having is obvious. But like recipe books did not turn everyone into Michelin chefs, it won't turn hacks into rigorous builders of trusted systems.
Bro, "AI" can't even program decently. It's still, to this day, worse than every human I've ever worked with. The idea that it's going to take over every industry is laughable.
no, those tools sucked in comparison to what ai is doing.
i just built a financial dashboard for my business. in realtime, i'm sucking in all the data from quickbooks, mercury, stripe, deel APIs and presenting it to myself, and my investors. it does all that sitting in cloudflare as a worker, for $5/month. took me a couple weeks of iteration, not a single line written.
from that i just told ai... "i'd like to get rid of quickbooks because yesterday it was down and i couldn't log into it. figure out what i'm using there and present me an interface that allows me to do whatever i'm doing in quickbooks, but with my existing data from stripe, deel and mercury."
it then dutifully went off and built me that app and now i'm running it in parallel and synchronized with quickbooks, until i feel that i can just turn that off. it is customized exactly to my business workflow and easily maintained with full unit tests. it costs me nothing but a $200/mo codex subscription and a few hours of work.
I was originally a Flash developer, when that died when Steve Jobs banned it from iPhones, I pivoted to frontend. I got to see web evolve over the next decade or so and loved every part of it.
Now, still as a frontend engineer albeit much more senior. The writing is on the wall, again. It's time to reskill for AI.
It's not bad news.
We as experienced engineers are needed to help build in the guardrails, frameworks, the new tooling, etc. There is a lot of work that is needed that requires actual engineers with technical experience.
Like the early internet, I see this as a new beginning and we get to decide how it evolves.
I think that is the nature of being a software engineer, even before AI you had to keep learning new libraries/SDKs/APIs/languages to keep up with the industry trend. I think it will remain the same in the age of AI, good engineers will adapt.
> Meanwhile, Cursor and Viget have blogged about migrating their codebases from Solid and Lit, respectively, to React.
That was maddening to read. Here I was thinking we should all be migrating to better performing frontend solutions. Does it really take an agent that many more tokens to write Lit or regular HTML and HTMX?
I can't wait for the asteroid to hit C-Suite and senior execs. These are the most meaningless jobs and easiest to automate, with the most cost savings.
There is going to come one day that companies (and any website owner) will wish they didn't use AI and instead used a simple (maybe even plain-text) website. That will come when people really need information from that website but can't get to it.
Absolutely this. Although I like the shiny and polished looks, getting the info out is the most important, and it is hard yo beat pure black-on-white text
I wish that was the answer to all of these people setting their spouse, their grandmother, their boss loose with LLMs to build sites. It should just be acceptable to have a simple site that conveys information well, and then we wouldn't need to boil the seas to generate pages of shiny slop.
> If you write a React component with an agent, though, then the risk of just yolo’ing it into production is (typically) much lower.
If your state management gets out of sync with the form controls your users are working with it can become pretty high risk because now users see one thing and something else actually gets sent to the backend.
Frontend development - perhaps more than any other kind of software development - is “what you see is what you get”. There aren’t a lot of ways a website can be invisibly broken.
Systems engineering isn’t like that. Vibe coders beware: A database could silently corrupt data, or have latent security vulnerabilities. It might have a badly designed query language that will be difficult to change later. But frontend code can be rewritten at will. It’s a perfect target for LLMs.
At the same time I have found LLMs far less competent at frontend because they don't seem to have any visual understanding of what a good website looks like.
"There aren’t a lot of ways a website can be invisibly broken."
??? Have you never debugged weird react state before??? Have you never used a nontrivial SPA before? Even the most simple react SPA has about a trillion states.
I don't really know how to respond to your statement than "no, they can definitely be invisibly broken."
React state bugs are usually not invisible failure modes.
Other kinds of software - for example, the backend of a website - have far more invisible failure modes. Security. Correctness. Performance. Bad API design leading to overfetching. Etc. Software works great on your machine and when you demo it, but it falls apart at scale. Endpoints that aren't secured properly. Race conditions cause silent data corruption. Memory leaks. And so on.
If claude messes up writing a react frontend, we're more likely to find the problem quickly. And frontend bugs usually don't continue to cause problems after they've been fixed. Silent bugs in the backend are more dangerous and more expensive.
When done properly, yes. But AI agents are doing a lot of data logic in the frontend that should be in the backend behind secure api's. When this happens, the frontend becomes a massive risk.
Maybe I'm an outlier here but I moved into front end to steer products and the resultant architecture into making sense from the incomplete or contradictory "requirements" we got from product and even UX. I've been doing this for over 20 years and every company has needed good engineers to make the product a logical reality.
LLMs piping out UIs from PM prompts aren't solving that imho.
I spent 2 years developing and polishing a front end web application, and then some other engineers at my company pointed Claude at it to try to reproduce it. The slop it crapped out would make most if not all our users leave for the competition. Then they had the nerve to ask me to explain what was bad about the sloppified version. I'm looking for a new job.
In my useless opinion, most people are not aware of how big AI is and how bigger it will be in 3~5 years. It's not worthy to be worried about it, because it will be a global problem, but there are people who are completely blind about it
Even if that is the case, the current models are already revolutionary. We would spend years discovering all the capabilities of these things. And that is considering that the plateau would last for years. And I didn't say that nobody is aware. There are some people that follows the breakthroughs and model releases (in my life), and they agree with me
I think they meant more along the lines of "This narrow problem today is not worth worrying about, because soon it will be a larger and bigger problem. "
"Note I’m not saying there are zero risks: the agent could mess up accessibility, it could cause an infinite loop that blocks users, etc. But in general, frontend code is a lot more ephemeral and replaceable than other types of code. So I expect many AI coders will feel comfortable just letting their agent handle it unsupervised (for better or worse)."
This is a weirdly reductive take on frontend correctness. Just for the record, I'm a backend dev. So I don't have much stake in this game.
This idea is, of course, not uncommon. "If the backend has to treat the frontend as adversarial anyway, and has all this cool stuff (constraints etc) for guaranteeing consistency of the system, then the frontend can just do whatever, right?" It plays into a lot of biases around typical frontend devs, typical backend devs, language stereotypes, etc. So it _sounds_ good.
Let me tell you for a moment about one of the spookiest bugs I've seen. It was an app for sorting personal photos. You'd upload pics/vids off your phone, they appear in the UI, you click a folder for them to go into (or click delete to discard), etc. Simple app, right?
Well, naturally, pics from even vaguely modern phones are regularly 5MB or more. Not really something you want to sling around while a user is browsing and their main activity is going to be looking at said pic to decide what folder it goes in (or if it gets deleted). So we thumbnail. And the main app only ever shows the user the thumbnails. The backend organized things quite simply: it gets a list of images from the frontend, it assigns each of them a zero-based index, and generates a thumbnail you'll also access via index. Imagine a URL scheme like `images/0` and `images/0/thumbnail` serving the real assets and the thumbnail.
Well, this app had a bug at one point. The backend was indexing by the arbitrary order the user uploaded them in. The frontend was mostly doing this too. Unfortunately the logic for thumbnails was incorrectly indexing by the "taken time" (which was a post-upload timestamp constructed by looking at basically every available timestamp and picking the "best" one. i.e. hopefully the one the iOS camera app adds, but obviously pics come from other places too and you never know what a user will upload). The end result being users would upload pics, see a thumbnail of an accidental pic they took of their shoe, hit delete. But actually they were deleting a pic of their baby or similar.
Literally none of the testing caught this for 2 main reasons: headless tests don't look at images, and you can't write an assertion like ("does this image look like a downscale of this other image") (at least not easily... i guess image models could do it now? but probabilistic? not a word i like in my unit tests? I digress, this predated the current crop of "AI").
Now let me generalize: your frontend isn't just a weird way to call RPCs on your backend. It's part of the application. I don't think you can just hand-wave. And as we saw above, you can't even say "well the frontend is stateless! any bug is 1 deploy away from fixing!" - deploying the frontend didn't get anyone their baby pictures back.
there seems to be at least one service that your LLM can find if it searches the internet (e.g. in my case "best rental API") that poisons your context into thinking that that one service is the definitive and only available API for that data
In a related but less public way, I've found myself less motivated to write internal documentation at work. It feels like in a lot of cases, it'd be better for the reader to throw Claude at the codebase to explain things.
A separate document explaining how a system works can go out of date. While I'm quite good at highlighting the most important insights and non-obvious traits of a system, I'd still be guessing what the reader needs from my doc at the end of the day. Claude can answer questions based on what the reader is curious about and wants to learn.
Some types of document are still useful, namely higher-level directional or philosophical topics about the intention of a codebase. e.g. "we would like to eventually move X system to Rust for Y reason", "we chose mutable data structures over immutable in this part of the system because Z". But fewer than before.
Yeah, I feel similarly, and kinda grieve how this is not longer a valued skill... I think I've cultivated my taste in many important social dimensions through thinking about what is important to leave in docs and in a note. It's a little strange to know that rarely will people do this work of honing this sense anymore...
>But in general, frontend code is a lot more ephemeral and replaceable than other types of code.
Tell that to Digg.
If you slop oit a front-end, your users will leave. It's that simple. The front-end is the part the users use the most, nobody cares how "elegant" your APIs are, or what kind of database the system uses. The front-end matters to users, a lot.
I've seen AI take a perfectly good front-end and slop out a piece of crap that will make the users leave in droves. And when nobody raises their hand (or in my case, everyone else is in the grip of AI psychosis) then bad times are ahead.
I offer this, as someone who is heavy into AI development (as a coding assistant, not vibe-coding). I was pretty fast before and now I'm easily an order of magnitude faster.
* I scrutinize every line of backend code and frequently make revisions.
* I pay just barely enough attention to the frontend code to make sure it's not doing anything dumb.
* I pay zero attention to the CSS.
Either our expectations for UI interaction are going to go up a lot, or FE devs should start looking at product management roles. I think the asteroid metaphor is apt.
As someone who loves FE the slop people are pushing where I work is really quite upsetting to me.
I reviewed some code earlier which clearly wasn't intended for consumed by humans and I know with certainty this dev wouldn't have understood 90% of what that code did. It looked like hours of Claude trying to force square into a round hole tbh, but it worked.
Was I going to reject the PR because of my code preferences? No, of course not. Ultimately it doesn't matter anymore. If Claude can maintain it then who cares... Even if it's broken (which it didn't seem to be) it's just FE anyway so not the end of the world.
I wish it did matter of course, but like most of coding these days, it just doesn't.
> Was I going to reject the PR because of my code preferences? No, of course not. Ultimately it doesn't matter anymore.
You absolutely should've rejected it. Putting slop into production is a huge trap, because now the humans can't understand it and the LLM is just going to make it worse and worse over time.
Being a programmer in the era of AI now means configuring the AI and run environments to produce the sought result, than manipulating the code itself. It's like how C runs on top of assembly.
It's nothing at all how like C runs on top of assembly. C has semantics. Code means something without ever being run. You can check in your source code and (assuming it actually follows the C standard and your compiler's documented extensions, which almost no real C code does - C was a horrible example) it will work just as well in a year as it does today.
Try checking in only your prompts and nothing else. Just the parts you actually typed. See how well it works to regenerate the same application next week, let alone next year. Prompts are fundamentally a different sort of thing from code. Do not mix them up.
> (assuming it actually follows the C standard and your compiler's documented extensions, which almost no real C code does - C was a horrible example) it will work just as well in a year as it does today.
> See how well it works to regenerate the same application next week, let alone next year. Prompts are fundamentally a different sort of thing from code
I mean... yes, but also no. C is actually a great example. So much of the code we wrote is about manipulating the specifics of that specific computer system we happen to be using at that exact moment. Everything from cpu specific instructions to how the ram behaves or how much of it there is all the way up to how library functions operate at any given point in time.
And in relatively short amounts of time, it can all change out from underneath you.
Being a programmer has always meant dealing with abstraction layers and climbing up the abstraction ladder.
You are simply learning a new language, whose syntax happens to look like English (or whatever language you speak), but with new, undiscovered, and constantly changing design patterns and best practices.
If you outsource everything to a llm, I just don’t think you get very good results right now. Language models aren’t great at remembering all the little design decisions that are needed in medium to large projects. The code ends up riddled with semi-conflicting design choices, which have been slammed together and maintained by context inertia. For a lot of projects this is more than fine - sometimes higher quality work simply isn’t worth my time. But for a lot of projects, you will pay for that slop later.
But I think there is a useful middle ground. If you carefully review all the decisions an llm makes, I think you can often be more productive than just programming everything by hand. But it’s a very different way to do engineering. I think I learn more this way than if I program by hand - if only because I’m touching more code. My mind roves around the design space a lot more. I don’t lock in as much as when I’m programming each piece individually.
If you work with an llm like this, your engineering skill matters more than ever.
But solving hard problems and developing systems intuition was. If you're trying to tell me that people working on code bases predominantly written by AI are still solving hard problems themselves, or understand the systems the AI is building, I think they're lying to themselves.
I work on a codebase with a lot of math and I've both learned and implemented a bunch of things that were beyond my level before I could iterate on it with AI. Same for binary reverse engineering. On the other fixing tedious bugs by hand and typing the code in is not something I miss at all.
From what I’ve seen they get to the level of understanding of systems way faster than we did because they iterate and experiment with this from day one. We had to spend years understanding code syntax, and other minutia.
Also, you need a different kind of toolset/skillset to grok a system that was vibe coded all the way - different failure scenarios. Older devs tend to just say “see, I told you, it’s just spaghetti underneath, you need to clean it up”, and the newer gen learns to work with that spaghetti.
I have a blazor app in poduction, Codex made errors which it could not fix and it required me reading documentation. I won't argue that it didn't make my life simpler, tasks which would take me days are done in hours, sometimes without touching code.
Yea, i still deal with a ton of code in AI heavy workflows. If anything the frustrating part is absorbing the code quickly enough. AI (Claude for me) writes in cryptic text and the code flows can often be non-obvious.
I need (and am exploring) custom review tooling to improve this AI->Human code flow. Reviewing PRs were always the hardest part for me in programming. They were often full of the developers decisions and you have to rediscover those as you're reading code for it to make sense[1]. However i find this even more difficult to discover these decisions from AI.
However unlike human PRs we can ask more of AI. Rarely have i had a developer put on a presentation for a PR - but AI could right? AI could produce a guided walkthrough of the code. Not sure if it will help of course, but my thought is we're all stuck in the old "PR review flow" but instead of PRs it's AI - and the volume of them is far greater than anything prior. So i expect we need to tweak how we review, how we get information from LLMs.
[1]: I'm speaking generally, and about larger PRs. Not some small func where you can easily see what it does. Business logic and complex code can be difficult to decipher in PRs, imo.
luckily there are still some people who know and care about actually understanding what is happening. but dwindling. market correction needed to wash larpers out
> For better or worse, React is heavily overrepresented in the training weights, and “agent experience” is starting to matter more than developer experience.
Another winner takes all situation. This and the other post on agent tool use[1] makes me think we are in a period of great homogenization. I'm sure there's a group that's delighted about this but for me, it just makes me feel a little uneasy.
[1] https://armature.tech/blog/which-tools-coding-agents-install
See https://arxiv.org/abs/2407.02209 but I hope researchers are looking to tackle this problem
> period of great homogenization
It's called alignment. And that's the correct word for it in my opinion.
The most important part is that we have to find a way to tackle the knowledge enshittification problem.
Because now that slopcode and slopposts are everywhere, the average quality level will decrease. We had the peak internet knowledge, and it's now gradually getting worse.
That's not a personal opinion, that's how attention matrixes and their keys/values computation works.
It was once called 'natural monopoly'
My wife's experience is managing teams and has been involved in creating corporate websites using contractors. She doesn't know anything about html at all. She just manages people.
She's now working on a biotech start-up pre-funding. They have pretty much zero money. They want to redesign their website.
I set her up with my deepseek key and she's been using flash-v4 with image detection (I think deepseek says it's still experimental). She just provides the current website, tells how she wants it to look and it works great. I mean it's obviously ai generated but she gets quick results and is able to tweek things to look better by just telling deepseek what to do. She was even able to change the text in one of their images to its new product name without having to contact the graphic designer.
This is a simple example. Basically it's one of those earlier jobs that you could get someone to bid on for $x but now you can do yourself for $20.
My mother has a psychology practice and wanted to update her 20 year old website.
I did the first few prompts (single static page, basically a brochure, don’t worry about hosting) into ChatGPT to get her started and a few days later she sent me a .zip which I uploaded.
Only took a while because she wasn’t paying for ChatGPT so had to wait for the free limit to reset quite a few times.
But overall it was super empowering for her.
Not that I'm in favor of spending money unnecessarily, but I find it kind of surprising she didn't just spend the $20, given this was a website for her professional money-making business.
> I find it kind of surprising she didn't just spend the $20
If it because the poster's mother dislikes the idea of subscriptions, then I totally get it. (Though I admit I have no idea if this has anything to do with the poster's mother's decision.)
I'll go out of my way as much as I can to avoid signing up for anything that is a subscription because of how horrible the entire corporate world has abused subscription dark patterns over the past couple of decades.
These days just the packaging of anything into a subscription makes me assume there is intent to screw people out of money.
While signing up for LLM usage is fairly low risk as far as subscription lock-in goes, someone non-technical might not be aware of that fact.
Does updating the website bring in additional money?
Presumably a nicer website could lead to more customers. Otherwise why update the website at all?
> Presumably a nicer website could lead to more customers.
It depends what "nicer" means. Paying a web designer/developer to create a bespoke website doesn't guarantee that said website will have a better UI/UX and, most importantly, "convert" at a higher rate. Not every web designer/developer understands IA, content, performance optimization, etc.
The popular AI tools produce websites that are clearly AI-produced if you know the patterns, but that doesn't mean they can't be usable and quite effective at converting. And if you have a bit of skill with prompting, AI can be really good at working through performance bottlenecks, etc.
Changing office address? New qualifications? Better copy?
so. if website does not lead to customers - why do any of that? actually why have website? Once you say yes to website you imply having one is somehow valuable, so in that case why not spend 20 dollars.
Just because websites can exist. Sometimes things are done for fun, or just for the sake of it.
This perspective declares all the personal websites of people worthless, however, they are priceless.
Because people's curiosity and creativity is priceless as well.
Similarly, why create art and share it around if it doesn't bring in customers? Why make music, take photos, dance? Do we live to serve customers, or do customers' existence serve us to live a better life?
Questions, questions...
Of course, those are all great reasons to make a website. But this website was for a woman's psychology practice. I don't imagine she made it for fun or to express her creativity. The purpose of a psychology practice's website is to tell people about your psychology practice, so they will either sign up ($) or find information they need more easily (higher satisfaction → $).
If OP's mother runs a donation-funded pro-bono psychology practice then you can translate the dollar signs to units of happiness or whatever (which I would imagine is her real goal regardless).
This person is spending her time to create the website (with ChatGPT's help). How much is that time worth? Of course when it's your business hourly pay is tricky, but I imagine she values her time highly enough that $20 is not worth very much of it.
I know an accountant, who has a respectable amount of customers and enough income to live comfortably. Their company has a website which only provides a logo, address and contact information, in a 23 line HTML file, complete with all the white space.
If their company can do well with this website, her psychology practice can do as well.
Some of the best services I got was from professionals with no or mundane online presence. Instead, their abilities and expertise moved them to the top in their respective fields, and the whole community knows how and when to reach them.
So, maybe a shiny website is not a hard-requirement for a good psychology practice, IDK.
So, the thing is, maybe this was her way to explore new things and decompress while doing some mundane thing from her perspective for free. Sometimes it's not that $20 which you can pay without any negative side-effects, but the desire to see how far the thing goes without the token, or just because you don't feel any urgency to do the task, so that $20 looks like an unnecessary waste.
I have been in both places.
Yes, presumably that's why she did not update the website for 20 years!
If once she's decided to update the website—surely it's worth more then $20 at that point? Otherwise, she probably shouldn't be spending any time on it herself either, given the opportunity cost.
Obviously she decided otherwise, I just find it kind of surprising.
Like, in general, I'm always kind of surprised at what people are and aren't willing to pay for. People will suffer through horrible full screen ads in mobile games instead of paying $5 one time to turn them off. And they'll do this on a phone that costs $1,000+. I don't know their financial situation, but I would buy a cheaper phone and get rid of the ads...
From what I have seen, people's perspective on value lies on a very wide spectrum. My mom won't buy a YouTube premium subscription, because she said she likes the ads. It's a kind of discovery platform for her. It's not that I share the same perspective, but I respect her about that.
I know people who won't subscribe to a service, but go through great pains to get the thing the service offers (another brand, similar yet inferior service, etc.).
For example, if an application I use has ads (i.e. a game), and it needs a subscription to remove these ads, I'll simply won't install it, or get the paid version of it from the start. For me, playing a game with no end and paying monthly to "hide" advertisements doesn't worth it.
These things are very subtle and are sometimes about ethos, and not about value at all. For instance:
* You propose I buy your ¤4 overhyped toothbrush. If I spend 10 minutes digging around, I could find a sub-¤1 toothbrush. Effectively I'm doing 10 minutes of work at less than minimum wage in order to avoid owning the premium toothbrush. Do I do this? I might! The reason is that I want no part in your hype.
* You offer me a loyalty scheme that will genuinely save me money if I participate. I only have to pay attention to your offers and routinely identify myself and I might get 10% savings, all the time. I decline this because you don't get to use my brain cycles for your profit, even if I profit too.
* You offer me a free thing that happens to track me and funnels data to data brokers. I decline it even though it does me no actual harm at all, because I don't want you to make money from personal facts about me, even when the facts are robotically aggregated and anonymised. Ethics are at stake: ethically speaking, you can just piss off. That is worth more than a free gadget, and it's not even an equation.
> but I imagine she values her time highly enough that $20 is not worth very much
Did you notice that this is completely unrelated to the idea that the $20 would turn a profit for the business?
It's just the argument "$20 isn't very much money, so why not spend it?".
> Otherwise why update the website at all?
Because you have it.
There is an entire industry built around this premise, yes.
The key is you're doing it yourself, so it takes your time away from something else. And if something doesn't quite work out, who's the backup?
For small shops and startups, this probably doesn't matter. For mission critical sites, who's more dependable? AI is a tool, that's all it is.
Every static HTML pages built this way with ai is another WordPress site that wont get built by contractors just to be neglected, compromised, and used to launch phishing and ransomware attacks.
Maybe, but what about when "Claude do this" picks WordPress for someone that doesn't know what a reverse shell is?
Still a win. Claude loves to write code from scratch instead of searching the web and pulling in dependencies, so you’ll likely end up with fewer garbage tier plugins written by people whose average skill level is put to shame by bootcamp grads; those people don’t know what a reverse shell is either.
Every static HTML pages built this way with ai is another WordPress site that wont get built by contractors just to be neglected, compromised, and used to launch phishing and ransomware attacks.
You mean the very web sites that the "AI" trained on?
If you think AI can't make buggy code, there are several dozen thousand comments on HN that disagree.
Ai cant make a static HTML page sitting on something like cloudflare pages have an rce bug.
She wouldn’t be able to do any of that without you setting it up first. She would not know anything about deepseek, meaning of a folder with index.html file. Would not know even what to ask ChatGPT, “I need a website”? You make it sound like easy task anyone can do, when you provided 99% of otherwise impossible for them to easily obtain key knowledge needed.
Any idiot can spend one minute now to acquire this amazing knowledge that you can just ask ChatGPT “I need a website”. If they don’t know now, sooner or later they’ll know from a short TikTok or something in the next couple of months/years. Not to mention AI companies are spending $$$ on ads to raise awareness; professionals already know they can create slides with LLMs, literally nothing’s different when it comes to most content websites. Yes, it’s an easy task any idiot can do. Head in sand, pretending it’s some sort of $10k for knowing where to hit situation isn’t going to reverse the trend.
You're not accounting for how deeply unpopular AI is with the general public. Would anyone be making those TikTok videos in the first place? Would OP's wife want to use AI if he wasn't there to advocate it?
Given that my nontechnical coworker has shared quite a few of these TikToks/Google Shorts with me, yes tons of people are making them. I don’t know what kind of alternative reality you live in, you can just go to any social media site and search for “create website with ChatGPT” etc. and check the view counts.
Plus the context here isn’t any random person, but people who intend to pay for website development for their businesses.
you can literally do that to chatgpt
Haha, yes. Can have it code a webserver from scratch on a disposable vape. Should be quite theraputic.
> She was even able to change the text in one of their images to its new product name without having to contact the graphic designer.
Is that legal... ? I admit I don't know much about graphic design is contracted.
Just like if you're a programmer the company paying you owns the work.
will not not necessarily be true for AI generated works in all territories
https://news.ycombinator.com/item?id=49382041
That's only true if the picture was specifically commissioned by the company.
If it's a stock image or a licensed image, they need a license to alter the original image.
It's generally a little more complicated than that. Not up to writing paragraphs about it so I'll just leave this as a note so anybody that's interested can do their own research.
show us these "corporate websites"
That was just part of her work. She was employee #11 at a now $4.5B biotech company so everyone that early on had to do everything. She worked for Novartis, for booz allen, and others. She usually hired McKinsey consultants for projects but the web stuff was for smaller companies. She managed a team in India for one of her companies that took care of the web stuff. Basically just look at any top-10 Pharma or government website and you might see what she managed.
I choose to imagine that you guys are like a cute "opposites attract" Marxist x capitalist power-couple :)
i started a web development business in 1995 and have been coding web stuff for decades.
like the OP, all those years of learning how it works under the covers gives me a distinct advantage since i can easily tell what is slop and what is going to perform. as well as how to get AI to iterate and fix the things it isn't good at (yet). i feel like now i have super powers. i love it and can't get enough.
the fact that your wife can be productive at lower cost means that more products will be built and more opportunities will be made. instead of spending money on website design, her business can spend money developing biotech.
i welcome that, even if it means that things will change from the past and those that have made a career out of charging thousands will get replaced by a $20/month subscription.
Would you say this is a similar but perhaps bigger change to frontpage/Dreamweaver/web app frameworks? Those all drastically cut down tub development time for web pages and at the same time opened up new possibilities. I wonder how it'll pan out with AI
FrontPage, Dreamweaber and co seemed to fell into obscurity when HTML5 & CSS3 came up and from that point web development went much more complicated than that.
Granted, it seems it is easier now to give prompts to an LLM so it will spit things out than back then where you had to know the tool and fiddle with menus and thingies to sort of try to produce a result.
But the drawback seems to be the same for both - lots of unreliable and unoptimized code that will go live if you don't know what they're doing under the hood and can potentially cause issues in the future. Remember that all this tools do are sorts of interpolation. I've seen lately lots of "modern" websites with tons layers of !important in their CSS and who knows the monstrosities that they have with their JS.
Meanwhile I and all people like me still don't have a job because all of this and the alleged "thousands" I was getting for doing this things.
the thing you haven't come to terms with is that it is more important to ship than it is to worry about the issues in the future. i thought RoR taught everyone that.
the beauty of what's happening today is that AI might produce slop on the first try, but you can also refactor just as easily too. "clean up all !important and optimize the code for speed", is just as easy of a prompt and it'll go off and do a fantastic job at that. that is what the OP was talking about in his article.
you don't have a job because you're not adapting to this new world order quickly enough. i predict that there will be insane demand for porting existing applications off these over priced SASS contracts and building tailored products for businesses. as much as they are trying to convince people, you don't need a $5m/yr salesforce / servicenow contract when you understand the business model and can envision a better solution to the actual problem out of thin air.
i'm burning 500m+ tokens a day on a $200/mo subscription. it is creating 10000x more value for me. it is no longer how much you know HTML/CSS, it is how well you know how to get AI to do what you want it to do and the only way to develop that skill, is to do it.
> you don't need a $5m/yr salesforce / servicenow contract when you understand the business model
I'm sure many people will think that; and I'm also fairly confident that 5 years later they'll take their obscure, bug-ridden, vibecoded monstrosity that nobody really understands, throw it out, and get a salesforce contract. Technical debt doesn't disappear just because you have agents working for you; in fact, it is exploding right now, and I would bet good money that 2-3 years for now there will be a flourishing cottage industry of consultancies specialized in dealing with such monstrosities.
Context is important in a response like this. Giacomo is a Sr. Presales consultant at a SaaS company that sells "The AI operating system for modern Finance." Of course you're adverse to things because it is a direct threat to what your business offers today.
If I can vibe code up a solution that fixes my problems for $200, it doesn't matter what happens in 5 years. The rate of change that's happening today in the increase in quality of the output of the models, I can just tell it to fix whatever we have and make it better.
Spoken with the true arrogance of someone who has no idea about the real world of finance.
When your software has to produce legal documents, satisfy auditors, and certify a bunch of things to shareholders and governments, I'm sorry but your $200 vibecoded hack is not going to cut it. You will have to get it certified, to get it audited, to get it secured according to this and that legislation, to prove where it gets each and every number and how it modifies it, and then you will have to maintain it because standards and laws change every other day. Sure, get an agent to cook you up a full consolidation solution, see how quickly you get laughed out of the room by grown-ups.
Btw, I am not "adversed to things" - the stuff we do has agents built-in, that do all sorts of wonderful things. The difference is that we can demonstrate that they do them right, with financial intelligence standardized and battle-tested across thousands of customers all over the world. Our customers don't have to worry about rewriting their pile of vibed hacks, or about being humiliated by auditors, because we take care of all that stuff for them. That's worth more than the pennies you save by having cousin' Vinnie vibe up a multidimensional financial engine that takes 3 days to calculate trillions of useless intersections before falling over.
You sound like someone who was adversed to cell phone cameras taking over film.
Get ready. AI is coming for every single industry, full steam.
> Get ready. AI is coming for every single industry, full steam.
It is, indeed. The C-Suites at every company are making plans: https://www.youtube.com/watch?v=DmU9uovmT2A
Hilarious!
But more real... Giacomo fails to take into account the concept of when OpenAI decides to create an audited financial service that guarantees that their output is valid. ChatGPT Work is already a beginning of that future.
That's like saying Google offering GMail ended webmail - there are still plenty of webmail services around: some better, some worse, some more specialised. The same will be true of most sectors.
AI is another tool in the toolbox, not magic. It's not about being pro or against it, it's here already (hey, we sell it!) and obviously it will continue to get better; but it won't magically solve everything for $2.
i've written smtp servers.
it has the best spam control on the planet.
i'd never use anything else.
That's great to know, Jon. I suspect non-US governments have a different opinion. Heck, even the US government likely has a different opinion. Not to mention Google competitors in any other sector.
one out of 10000000 customers.
Have you any idea of how many businesses need a financial consolidation tool, out of the total number of businesses out there?
No, of course you don't, you write email servers.
Sure, your local garage will now do their accounting in vibed apps instead of vibed Excel files. I suspect multinationals will choose not to.
I can see I'm under your skin as you're trying to make this personal. I have a long history of being early and correct with many things. What is transforming under your nose, faster than anything ever before, will take you for surprise, unless make a conscious decsion to wake up. Do yourself a favor and listen to what he's saying... https://youtu.be/NYFGCESmikA?t=2883
I was not the one who brought personal qualifications into a public argument, Jon. I guess you don't like it when your lack of specialized knowledge becomes evident.
As I said, I already know what AI is doing - I already use it just fine, I even help building and selling it. The impact it is having is obvious. But like recipe books did not turn everyone into Michelin chefs, it won't turn hacks into rigorous builders of trusted systems.
So you think AI will only ever be a recipe book? That's the disconnect we have in opinion.
Bro, "AI" can't even program decently. It's still, to this day, worse than every human I've ever worked with. The idea that it's going to take over every industry is laughable.
it is funny having some anon tell me something that is 100% different from my own experience and i've been coding for 30+ years now.
no, those tools sucked in comparison to what ai is doing.
i just built a financial dashboard for my business. in realtime, i'm sucking in all the data from quickbooks, mercury, stripe, deel APIs and presenting it to myself, and my investors. it does all that sitting in cloudflare as a worker, for $5/month. took me a couple weeks of iteration, not a single line written.
from that i just told ai... "i'd like to get rid of quickbooks because yesterday it was down and i couldn't log into it. figure out what i'm using there and present me an interface that allows me to do whatever i'm doing in quickbooks, but with my existing data from stripe, deel and mercury."
it then dutifully went off and built me that app and now i'm running it in parallel and synchronized with quickbooks, until i feel that i can just turn that off. it is customized exactly to my business workflow and easily maintained with full unit tests. it costs me nothing but a $200/mo codex subscription and a few hours of work.
I was originally a Flash developer, when that died when Steve Jobs banned it from iPhones, I pivoted to frontend. I got to see web evolve over the next decade or so and loved every part of it.
Now, still as a frontend engineer albeit much more senior. The writing is on the wall, again. It's time to reskill for AI.
It's not bad news.
We as experienced engineers are needed to help build in the guardrails, frameworks, the new tooling, etc. There is a lot of work that is needed that requires actual engineers with technical experience.
Like the early internet, I see this as a new beginning and we get to decide how it evolves.
I think that is the nature of being a software engineer, even before AI you had to keep learning new libraries/SDKs/APIs/languages to keep up with the industry trend. I think it will remain the same in the age of AI, good engineers will adapt.
> Meanwhile, Cursor and Viget have blogged about migrating their codebases from Solid and Lit, respectively, to React.
That was maddening to read. Here I was thinking we should all be migrating to better performing frontend solutions. Does it really take an agent that many more tokens to write Lit or regular HTML and HTMX?
At least last year most agents wrote materially worse code in htmx.
> Does it really take an agent that many more tokens to write Lit or regular HTML and HTMX?
Definitely not. That's skill issues. The shadow DOM can be a bit tricky.
I can't wait for the asteroid to hit C-Suite and senior execs. These are the most meaningless jobs and easiest to automate, with the most cost savings.
There is going to come one day that companies (and any website owner) will wish they didn't use AI and instead used a simple (maybe even plain-text) website. That will come when people really need information from that website but can't get to it.
https://sparkbox.com/foundry/helene_and_mobile_web_performan...
Discussion regarding that post: https://news.ycombinator.com/item?id=46494734
Absolutely this. Although I like the shiny and polished looks, getting the info out is the most important, and it is hard yo beat pure black-on-white text
I wish that was the answer to all of these people setting their spouse, their grandmother, their boss loose with LLMs to build sites. It should just be acceptable to have a simple site that conveys information well, and then we wouldn't need to boil the seas to generate pages of shiny slop.
> If you write a React component with an agent, though, then the risk of just yolo’ing it into production is (typically) much lower.
If your state management gets out of sync with the form controls your users are working with it can become pretty high risk because now users see one thing and something else actually gets sent to the backend.
If your coding agent screwed up so bad that wrong data is being shown to the user, in August 2026, thats more on the prompter than anything else.
> The frontend is less risky to just hand to an agent.
The smaller the problem, the less risky it is to do this, regardless of the nature of the problem. This is not a frontend-specific issue today.
Frontend development - perhaps more than any other kind of software development - is “what you see is what you get”. There aren’t a lot of ways a website can be invisibly broken.
Systems engineering isn’t like that. Vibe coders beware: A database could silently corrupt data, or have latent security vulnerabilities. It might have a badly designed query language that will be difficult to change later. But frontend code can be rewritten at will. It’s a perfect target for LLMs.
At the same time I have found LLMs far less competent at frontend because they don't seem to have any visual understanding of what a good website looks like.
The UX is what makes the users use the apps, systems, games, their phones.
"There aren’t a lot of ways a website can be invisibly broken."
??? Have you never debugged weird react state before??? Have you never used a nontrivial SPA before? Even the most simple react SPA has about a trillion states.
I don't really know how to respond to your statement than "no, they can definitely be invisibly broken."
React state bugs are usually not invisible failure modes.
Other kinds of software - for example, the backend of a website - have far more invisible failure modes. Security. Correctness. Performance. Bad API design leading to overfetching. Etc. Software works great on your machine and when you demo it, but it falls apart at scale. Endpoints that aren't secured properly. Race conditions cause silent data corruption. Memory leaks. And so on.
If claude messes up writing a react frontend, we're more likely to find the problem quickly. And frontend bugs usually don't continue to cause problems after they've been fixed. Silent bugs in the backend are more dangerous and more expensive.
No need to engage with bad faith premises. You cant convince someone who is invested in their own FUD.
When done properly, yes. But AI agents are doing a lot of data logic in the frontend that should be in the backend behind secure api's. When this happens, the frontend becomes a massive risk.
Maybe I'm an outlier here but I moved into front end to steer products and the resultant architecture into making sense from the incomplete or contradictory "requirements" we got from product and even UX. I've been doing this for over 20 years and every company has needed good engineers to make the product a logical reality.
LLMs piping out UIs from PM prompts aren't solving that imho.
I spent 2 years developing and polishing a front end web application, and then some other engineers at my company pointed Claude at it to try to reproduce it. The slop it crapped out would make most if not all our users leave for the competition. Then they had the nerve to ask me to explain what was bad about the sloppified version. I'm looking for a new job.
In my useless opinion, most people are not aware of how big AI is and how bigger it will be in 3~5 years. It's not worthy to be worried about it, because it will be a global problem, but there are people who are completely blind about it
Nobody is aware, but neither are you. We may as well have plateaued already. You don't know that.
Even if that is the case, the current models are already revolutionary. We would spend years discovering all the capabilities of these things. And that is considering that the plateau would last for years. And I didn't say that nobody is aware. There are some people that follows the breakthroughs and model releases (in my life), and they agree with me
> It's not worthy to be worried about it, because it will be a global problem
Like hunger and diseases are global problems not to be worried about.
I think they meant more along the lines of "This narrow problem today is not worth worrying about, because soon it will be a larger and bigger problem. "
"Note I’m not saying there are zero risks: the agent could mess up accessibility, it could cause an infinite loop that blocks users, etc. But in general, frontend code is a lot more ephemeral and replaceable than other types of code. So I expect many AI coders will feel comfortable just letting their agent handle it unsupervised (for better or worse)."
This is a weirdly reductive take on frontend correctness. Just for the record, I'm a backend dev. So I don't have much stake in this game.
This idea is, of course, not uncommon. "If the backend has to treat the frontend as adversarial anyway, and has all this cool stuff (constraints etc) for guaranteeing consistency of the system, then the frontend can just do whatever, right?" It plays into a lot of biases around typical frontend devs, typical backend devs, language stereotypes, etc. So it _sounds_ good.
Let me tell you for a moment about one of the spookiest bugs I've seen. It was an app for sorting personal photos. You'd upload pics/vids off your phone, they appear in the UI, you click a folder for them to go into (or click delete to discard), etc. Simple app, right?
Well, naturally, pics from even vaguely modern phones are regularly 5MB or more. Not really something you want to sling around while a user is browsing and their main activity is going to be looking at said pic to decide what folder it goes in (or if it gets deleted). So we thumbnail. And the main app only ever shows the user the thumbnails. The backend organized things quite simply: it gets a list of images from the frontend, it assigns each of them a zero-based index, and generates a thumbnail you'll also access via index. Imagine a URL scheme like `images/0` and `images/0/thumbnail` serving the real assets and the thumbnail.
Well, this app had a bug at one point. The backend was indexing by the arbitrary order the user uploaded them in. The frontend was mostly doing this too. Unfortunately the logic for thumbnails was incorrectly indexing by the "taken time" (which was a post-upload timestamp constructed by looking at basically every available timestamp and picking the "best" one. i.e. hopefully the one the iOS camera app adds, but obviously pics come from other places too and you never know what a user will upload). The end result being users would upload pics, see a thumbnail of an accidental pic they took of their shoe, hit delete. But actually they were deleting a pic of their baby or similar.
Literally none of the testing caught this for 2 main reasons: headless tests don't look at images, and you can't write an assertion like ("does this image look like a downscale of this other image") (at least not easily... i guess image models could do it now? but probabilistic? not a word i like in my unit tests? I digress, this predated the current crop of "AI").
Now let me generalize: your frontend isn't just a weird way to call RPCs on your backend. It's part of the application. I don't think you can just hand-wave. And as we saw above, you can't even say "well the frontend is stateless! any bug is 1 deploy away from fixing!" - deploying the frontend didn't get anyone their baby pictures back.
It already slurped "the internet". You can't seriously say it doesn't count!
there seems to be at least one service that your LLM can find if it searches the internet (e.g. in my case "best rental API") that poisons your context into thinking that that one service is the definitive and only available API for that data
In a related but less public way, I've found myself less motivated to write internal documentation at work. It feels like in a lot of cases, it'd be better for the reader to throw Claude at the codebase to explain things.
A separate document explaining how a system works can go out of date. While I'm quite good at highlighting the most important insights and non-obvious traits of a system, I'd still be guessing what the reader needs from my doc at the end of the day. Claude can answer questions based on what the reader is curious about and wants to learn.
Some types of document are still useful, namely higher-level directional or philosophical topics about the intention of a codebase. e.g. "we would like to eventually move X system to Rust for Y reason", "we chose mutable data structures over immutable in this part of the system because Z". But fewer than before.
Yeah, I feel similarly, and kinda grieve how this is not longer a valued skill... I think I've cultivated my taste in many important social dimensions through thinking about what is important to leave in docs and in a note. It's a little strange to know that rarely will people do this work of honing this sense anymore...
> A separate document explaining how a system works can go out of date.
Good documentation does not explain "the how", it captures "the why."
Even the latter can be done automatically from meeting notes, tickets etc.
>But in general, frontend code is a lot more ephemeral and replaceable than other types of code.
Tell that to Digg.
If you slop oit a front-end, your users will leave. It's that simple. The front-end is the part the users use the most, nobody cares how "elegant" your APIs are, or what kind of database the system uses. The front-end matters to users, a lot.
I've seen AI take a perfectly good front-end and slop out a piece of crap that will make the users leave in droves. And when nobody raises their hand (or in my case, everyone else is in the grip of AI psychosis) then bad times are ahead.
I offer this, as someone who is heavy into AI development (as a coding assistant, not vibe-coding). I was pretty fast before and now I'm easily an order of magnitude faster.
* I scrutinize every line of backend code and frequently make revisions.
* I pay just barely enough attention to the frontend code to make sure it's not doing anything dumb.
* I pay zero attention to the CSS.
Either our expectations for UI interaction are going to go up a lot, or FE devs should start looking at product management roles. I think the asteroid metaphor is apt.
As someone who loves FE the slop people are pushing where I work is really quite upsetting to me.
I reviewed some code earlier which clearly wasn't intended for consumed by humans and I know with certainty this dev wouldn't have understood 90% of what that code did. It looked like hours of Claude trying to force square into a round hole tbh, but it worked.
Was I going to reject the PR because of my code preferences? No, of course not. Ultimately it doesn't matter anymore. If Claude can maintain it then who cares... Even if it's broken (which it didn't seem to be) it's just FE anyway so not the end of the world.
I wish it did matter of course, but like most of coding these days, it just doesn't.
> Was I going to reject the PR because of my code preferences? No, of course not. Ultimately it doesn't matter anymore.
You absolutely should've rejected it. Putting slop into production is a huge trap, because now the humans can't understand it and the LLM is just going to make it worse and worse over time.
I think what OP is saying is that the fact that a human can't understand it won't matter as the code will only ever be read by AI.
Being a programmer in the era of AI now means configuring the AI and run environments to produce the sought result, than manipulating the code itself. It's like how C runs on top of assembly.
It's nothing at all how like C runs on top of assembly. C has semantics. Code means something without ever being run. You can check in your source code and (assuming it actually follows the C standard and your compiler's documented extensions, which almost no real C code does - C was a horrible example) it will work just as well in a year as it does today.
Try checking in only your prompts and nothing else. Just the parts you actually typed. See how well it works to regenerate the same application next week, let alone next year. Prompts are fundamentally a different sort of thing from code. Do not mix them up.
> (assuming it actually follows the C standard and your compiler's documented extensions, which almost no real C code does - C was a horrible example) it will work just as well in a year as it does today.
> See how well it works to regenerate the same application next week, let alone next year. Prompts are fundamentally a different sort of thing from code
I mean... yes, but also no. C is actually a great example. So much of the code we wrote is about manipulating the specifics of that specific computer system we happen to be using at that exact moment. Everything from cpu specific instructions to how the ram behaves or how much of it there is all the way up to how library functions operate at any given point in time.
And in relatively short amounts of time, it can all change out from underneath you.
Being a programmer in the era of AI means not being a programmer
Being a programmer has always meant dealing with abstraction layers and climbing up the abstraction ladder.
You are simply learning a new language, whose syntax happens to look like English (or whatever language you speak), but with new, undiscovered, and constantly changing design patterns and best practices.
It depends how much you outsource to the llm.
If you outsource everything to a llm, I just don’t think you get very good results right now. Language models aren’t great at remembering all the little design decisions that are needed in medium to large projects. The code ends up riddled with semi-conflicting design choices, which have been slammed together and maintained by context inertia. For a lot of projects this is more than fine - sometimes higher quality work simply isn’t worth my time. But for a lot of projects, you will pay for that slop later.
But I think there is a useful middle ground. If you carefully review all the decisions an llm makes, I think you can often be more productive than just programming everything by hand. But it’s a very different way to do engineering. I think I learn more this way than if I program by hand - if only because I’m touching more code. My mind roves around the design space a lot more. I don’t lock in as much as when I’m programming each piece individually.
If you work with an llm like this, your engineering skill matters more than ever.
Typing the code was never the interesting part.
>> Typing the code was never the interesting part.
Code is fascinating! An abstract language to compile to program a CPU. So many ways to do this!
Now we're using natural language, which at times is extremely verbose. "[{}]" versus "array of objects", for example.
For me it definitely was
>Typing the code was never the interesting part.
But solving hard problems and developing systems intuition was. If you're trying to tell me that people working on code bases predominantly written by AI are still solving hard problems themselves, or understand the systems the AI is building, I think they're lying to themselves.
I work on a codebase with a lot of math and I've both learned and implemented a bunch of things that were beyond my level before I could iterate on it with AI. Same for binary reverse engineering. On the other fixing tedious bugs by hand and typing the code in is not something I miss at all.
From what I’ve seen they get to the level of understanding of systems way faster than we did because they iterate and experiment with this from day one. We had to spend years understanding code syntax, and other minutia.
Also, you need a different kind of toolset/skillset to grok a system that was vibe coded all the way - different failure scenarios. Older devs tend to just say “see, I told you, it’s just spaghetti underneath, you need to clean it up”, and the newer gen learns to work with that spaghetti.
I have a blazor app in poduction, Codex made errors which it could not fix and it required me reading documentation. I won't argue that it didn't make my life simpler, tasks which would take me days are done in hours, sometimes without touching code.
Yea, i still deal with a ton of code in AI heavy workflows. If anything the frustrating part is absorbing the code quickly enough. AI (Claude for me) writes in cryptic text and the code flows can often be non-obvious.
I need (and am exploring) custom review tooling to improve this AI->Human code flow. Reviewing PRs were always the hardest part for me in programming. They were often full of the developers decisions and you have to rediscover those as you're reading code for it to make sense[1]. However i find this even more difficult to discover these decisions from AI.
However unlike human PRs we can ask more of AI. Rarely have i had a developer put on a presentation for a PR - but AI could right? AI could produce a guided walkthrough of the code. Not sure if it will help of course, but my thought is we're all stuck in the old "PR review flow" but instead of PRs it's AI - and the volume of them is far greater than anything prior. So i expect we need to tweak how we review, how we get information from LLMs.
[1]: I'm speaking generally, and about larger PRs. Not some small func where you can easily see what it does. Business logic and complex code can be difficult to decipher in PRs, imo.
That is why I am very thankful for my job. We are not really allowed to use LLMs for much of anything.
That sounds like an absolute dream.
thanks i hate it
luckily there are still some people who know and care about actually understanding what is happening. but dwindling. market correction needed to wash larpers out
writing about your dates with an ai prompt are not that interesting