OP, you accidentally invented the design process that they teach in design school. You can officially call yourself a designer.

1) have a high level functional goal

2) be inspired and think of a motif

3) go wide within that motif and explore

4) narrow in on a single direction and iterate

5) fine tune the details until you are creatively satisfied and the design meets your goals from step 1.

Yeah, and the trade-off here is that it takes time. In the same way, you can write insightful blog posts or work emails using AI if you maintain creative control, but all of sudden, you're not moving at 100x speed. You're moving a lot closer to human speed.

This trade-off is something I like the least about AI. In the time it takes someone to produce excellent results using AI without giving up control, people unconcerned with quality or style can produce 100x as much. So by the end of the day, your inbox or your social media feed is mostly slop.

The fact it takes time is not a trade-off. It is an actual design work.

You forget that in any given blog post, work is bad, unless it's the kind of work the blog author does, then it's valuable

I mean, one part of the trade-off is that you can be a designer without having gone to design school. If you can use the tools you can iterate on the averageness of all these professions and fill all those shoes.

Choose 1:

A UX-designer with that outputs design with AI at the speed of a UX-designer without AI

A programmer that can output designs with AI at the same speed, and make your database, and make your helpdesk, and make your social media, booking systems, etc.

AI can do an awesome job at UI and come up with really nice look but I feel like there’s a lot of psychology behind UX that AI cannot necessarily pick up on the human element.

Explain what cannot be picked up on and why.

The LLMs have ingested courses and books on UI, UX, psychology, neuroscience, consumer neuroscience, etc. What makes it incapable of getting this information?

Ugh, my little kid talks to me like this.

Let's try again. What's a nicer way of saying that?

Clicking through the website with computer use isn't currently part of the website generation chain for most models, and while there are gaze prediction models, to my knowledge none are productized or universal to the point where you can chuck them in and consistently predict what draws attention on any given page

(Generally I think it is alright to interpret "can't do" in terms of "the common workflow that current models will afford" and not "definitionally incapable of")

Humans have the same limitations in this, we model an idea using prior knowledge for what will draw attention, then we test it and evaluate it.

I was not making an argument about people, I was trying to identify specific reasons why models can produce samey UIs unless guided

I have never heard this, but I like it a lot.

That sounds like basic project management

I'm sure this is all in the training data.

I like hearing it articulated for me.

It’s funny, this is exactly how I used opus to design an iOS app I’m building. Unfortunately fable was released and it seems my taste lines up with fable’s training quite a bit[1]. Now my app looks like ai anyway. You win some you lose some I guess.

[1] https://onda-cast.github.io/onda/

Before everything looked like bootstrap or one of the other 5 design systems. Website? Hero Image, 3 Colums, footer. This was peak design half a decade ago.

I am deeply offended by the bootstrap comparison, I was more of a Milligram guy in the 2010s.

I love the list, but the design process I taught in school was way different — it was a loop of two steps, started anywhere:

1. Do something. (Many of the five above seem nice! But there’s lots of methods that fit different tasks)

2. Ask some users please for the love of god! Show it to somebody! Get your mom’s feedback! Anybody!!!

Etc. It was drilled into me pretty hard at GaTech by their HCI prof, but perhaps he spent so much time reiterating it because it’s heterodox.

It doesn’t help that “design” covers everything from splash pages that sell SaaS to doors and blankets and stuff. I’m a little dubious of any single unified framework other than the aforementioned Frantic Dyad that could hope to cover all that space.

I think, many of the folks commenting below, mentioning that the OP hasn't actually studied in a design program in school, also fail to mention that the attorney, and later U.S. President, Abraham Lincoln also had no formal education in learning to read.

The latter, in his path toward literacy, and eventually a J.D. (That's what attorneys are nowadays anyway) was largely due to his stubborn autodidactism, the same mettle that the OP himself chronicles and showcases in his blog.

If you put most people participating (or lurking) here on Hacker News through this scrutiny, you really shouldn't be surprised that this is generally the case with anyone who's been working in data processing, Management Information Systems, Information Technology, or WhatEv we're going to be calling it in the next decade.

Nowadays, when I think of n00bs, I don't think about the computer illiterate, I think about the motivated neophytes who spent two years grinding along in an IT track in college at the snails pace of semester courses and now they can consider themselves entry level python programmers with an extra semester where they took "Introduction to Java - CS-107".

If you're going to survive in this industry, you already have an insatiable curiosity for everything under the hood and morbid obsessions with break/fix - emphasis on 'fix' :)

If it looks like Yakko (the OP of this thread) went through two semesters (that really is a freakin' year in trad educational systems) of design school, but in reality really didn't... You would be wrong. He just did it on his own and at a pace he was comfortable with, arriving at the same dock to berth his ship as someone who actually spent a year being forced learn it at a snails pace.

Not everyone can be a network engineer, or a systems administrator, or full stack developer, etc., but those who thrive upon the work are almost always exceedingly autodidacts in those disciplines...

You literally cannot keep up with the pace of evolving technologies if you're doing so in a legacy educational institution.

Interview question: "Tell me about a nightmare scenario you were faced with when you were without the knowledge to fix the problem and yet, did so anyway - how and what did you do to rise to the occasion, and how did that make you feel like when everyone was back online again?"

    Where ignorance is bliss, / 'Tis folly to be wise

I have such an insatiable curiosity, but it really strips the "magic" from the world more often than not. I think it's much more fun to make believe that no one understands LLMs, for example, than to know that not only do we understand them quite well enough, it is also quite boring in the end. As are all things once the curtain has been peeked.

The reason I added that quote, by the way, was because although nowadays people typically use "ignorance is bliss" as a blanket justification for willful ignorance, intellectual laziness, or avoiding uncomfortable truths, when English poet Thomas Gray coined the phrase in his 1742 poem (Ode on a Distant Prospect of Eton College), he was expressing a sentiment about the inevitable loss of childhood innocence.

    Yet ah! why should they know their fate?
    Since sorrow never comes too late,
    And happiness too swiftly flies.
    Thought would destroy their paradise.
    No more; where ignorance is bliss,
    ’Tis folly to be wise.

And yes, I knew I'd be downvoted for it.

    To each his suff'rings: all are men,
    Condemn'd alike to groan

> If you're going to survive in this industry

I really feel like it's very much random.

When job hunting -> does the recruiter shortlist you?

When on the job -> do you get laid off?

Neither has anything to do with your skills at the actual job.

> do you get laid off?

> Neither has anything to do with your skills at the actual job.

I hate to break it to you. But it definitely does...

it doesnt really. some layoffs yes, others no.

the high performers in a whole org that gets cut, still gets cut.

even the performance of the org itself doesnt matter if the business decides the opportunity cost of not doing something else is too expensive

A designer did make it! The process you undertook is called design!

We need to make ppl understand this!

"chat read this long ass tweet and do the same stuff make no mistakes"

Literally yes

> Most times I'll grab ideas from multiple places, combine them, and endlessly iterate on them until I get to something I'm happy with.

The "designer" happens to not to be human. At that point it just sounds like going to a website template store and scrolling endlessly through options until you find a template you'd say "meh, this will do".

That’s not how it works.

Iteration is a critical part of good design-work, no matter the tool you use to do it. You don’t have to be drawing things by hand to be designing them.

I had this out with one of my employees, they were giving me shit for using Claude Design, and just how many iterations it took to get what I wanted.

I explained that exactly what design is I just lack the skills to do it in illustrator and Photoshop but yes, no one gets it right the first time.

Best tip in here: when tweaking details like colors and fonts, just ask Claude to make you throwaway picker tools right in the site.

This is totally obvious in retrospect but I never thought of it (probably because of 25 years of thinking about tools in terms of reusable components).

Thanks for writing!

There’s a nice Apple session from this year’s WWDC about this where they suggest using the new resizing feature of the simulator (born to support the foldable iPhone) to have a “drawer” for all the controls and tools to fine tune your UX. I’ll be stealing that for the web as well

It looks like a lot of their early attemps are early gpt outputs or claude design outputs, which have a lot of the same smell. People are discounting how good image->html flows are these days though.

I had gemini describe this webpage as prompts, then fed it into an image -> code pipeline. Note that this is with 0 input from me:

https://html.non.io/railcode-example

This is with Opus 5.5 building it out on low effort, and yes the spec was already made from the original site, but 0->1 that doesn't result in purple gradients or claude brown is very much possible now.

Honestly looks better than the original

Disagree, the colors are harsh

The green is too bright but the animations are way better

The animation is more dense and busy. Not sure I'd qualify it as better necessarily.

But this thread is a bit like the "I could have drawn that" comment you hear from Average Joe saying about a Mondrian painting. Yes, you can mimic something (with or without AI) but you wouldn't have made _this_.

Don't get me wrong the 0->1 copycat site is impressive, but it lacks a lot of attention to detail the original has.

The art and engineer snob voice in me is also causing a bit of cognitive dissonance about this though. C'est la vie, funny times we live in

The original animation has people being run over by the roller coaster train... And this one's movement attempts to approximate real physics, the original doesn't. It was a lack of attention to detail in the original animations that stood out the most to me...

Ah, I meant the site design as a whole when I talked about attention to detail, not the animation specifically

As for the animations...

The second animation is a bit more... High fidelity in some areas? I don't know what to even call it. The clash of different pixel art styles is what I find jarring[^1].

But I guess my point was that neither animation is even _good_... Like both of them barely pass a cursory glance; there's lot of nonsense going on in both of them, that wouldn't get past any animator. So, I wouldn't say one is better or worse; they just different

[^1]: Edit for context, I used and even made _a lot_ of pixel art in primary school and my teens lol

I agree the original author's post is better than this. I think the point I'm trying to make is that "vibecoded" landing pages can actually be quite high quality these days, and the author was sandbagging a bit for their point that 0->1 results are bad (which fair, I'm guilty of the same).

They are correct though that these days you can navigate away from the slop pages that gpt 5 put out pretty effectively, and land on something that really is quite decent. The page I made is a great starting point for 60s of effort on my part.

Answer: because a human designed it in an iterative process. LLMs just wrote the code.

And it still carries some of the cliche features of LLM-'designed' sites.

It definitely does. Particularly under the hero imo.

Doing this is really time-consuming so certainly there were some corners cut -- need to also focus on product after all!

Particularly after we changed a lot of the messaging (after this post was written) I slacked off a bit. Keen to hear what's most annoying to you though, always looking for feedback.

The way the body content animates at page load is pretty annoying.

like what?

The website is simple enough that any design decicion could just as well ahve been done by a real human.

The em-dashes, and the gradients were always egregious because they're strong opinionated decisions. This is just a straighforward, clean, tasteful website (really what most sites should aim for).

it feels vibecoded though, it still has LLMisms

for sure

Yes, if the designer was named Claude or Astra, then it does indeed look like that designer created it.

I love this because, it reminds me that when people hate AI, They hate low effort AI slop.

This website demonstrates a tremendous amount of human effort, and it walks through the iterative design process that led to the final result. Taking inspiration, trying out ideas, tinkering—that's design.

It's uhhh. I don't want to uh... A "tremendous" amount of human effort?

You know, it's okay.

So because of the article's premise, I'm going to be a bit of a dickhead. I hope nobody takes this the wrong way, because it's not a judgement on the developers. I have also shipped way worse vibecoded frontends and this is coming from a UI guy, so it's not about being an AI hater or anything.

Your site looks like an AI made it.

Now there is an actually easy fix: just replace the glass panels and tabs with something that actually fits the rest of the aesthetic. And then I could buy that it was human-designed, at which point, it basically was. And just AI-coded.

Otherwise it's quite cool, and well done.

This makes sense yeah, I guess following a similar approach used on the buttons for the tabs as well. Point taken!

Yeah that's the only parts of the design that I could identify as being clearly AI generated, or at least is ai-coded, if it was human designed. Including the panels for demo, pricing, etc.

The rest of the design kind of reminds me of stardew valley's site, which is quite cute.

Hope you don't take this as evil criticism, overall I'm a big fan.

edit: it's not even AI coded, more like incohesive, and with the context of the article...but a simple fix and overall well done ;)

Not at all! I agree with what you said

> just replace the glass panels and tabs with something that actually fits the rest of the aesthetic.

Where is this glass?

Is it the lighter green cards on the home page? That's not really glass, that's just a different a lighter color scheme. The tabs also look fine.

i'm not going to argue with you about aesthetics. taste and all that. but to me there are clearly divergent aesthetics with the rest of the site, and some elements. my suggestion is to just align them, and it will be more cohesive.

This is the kind of creative work that I think will always need to be guided by human taste and vision but can be vastly enabled by AI. I really hope to see more people using AI to improve quality instead of velocity.

Cute but why are the people on the coaster standing and why are there people standing on the tracks

The people on the coaster aren't sitting out the revolution. The people on the tracks are standing in the way of progress.

a dime a dozen

26 info warnings though I'm impressed there are no up front errors but the warnings are of potential problems. It's the same blunder I see many people who don't read the standards don't understand or bother with until it actually becomes a problem. But AI learned from them. That's why the problem exists within AI, too.

https://validator.w3.org/nu/?doc=https%3A%2F%2Frailcode.dev%...

Turns out you you just need taste and the time to iterate

Oof, an unbanked turn right after the first drop. Claude must be a big fan of lateral g's.

What a jerk.

Can someone translate this for us simpletons who don't understand?

Go play roller coaster tycoon for a few days

this is awkward - the site definitely still looks vibe coded / LLMs helped make it

Looking at the source i was not surprised. A shell html page, that uses a shittonne of javascript, and obviously does not even include the static content per page load. Its sad that this is the status quo of the AI slop you see.

I have seen and paid expensive designers that did a much worse job

Well there's your problem, you're supposed to pay coders to do the coding not designers

You're proving that even with all the tools at our disposal now, taste really does matter. I love your website!

Even when building with AI, I find it very useful to be picky about the details. Sometimes picky enough to go down to the code.

I feel like people who get the most out of AI tools are the ones who actually use it as a tool, and don't treat AI as the decision maker.

I had a similar experience with my landing page for downread.com. I don't know whether the design looks vibe-coded, but I don't think anyone would put that much effort into a hand-coded homepage: the landing page (as well as the feature page) uses interactive, partially animated 'screenshots' that simulate aspects of the app. All buttons either do something or open a pop-up explaining what the button does.

That is a fun website design. My current strategy is to have a sub agent pretend to be a hackernews reader who hates AI slop. It then goes through the whole UI and removes all the slop.

It works really well, even with models like DeepSeek which have been completely poisoned with claudeslop. You get a website that actually makes sense, and doesn’t have random slop lines like “Logout— it stays logged in until the user logs out”

Now I guess I need to do a pass with a whimsy subagent to add fun little elements like this.

I got inspired by this post to add a sushi belt to my Japanese learning website: tokidoki.meertens.dev

Whether it's actual creativity or not, it is a lot of fun to have an idea and see it executed immediately!

Loved the process write-up and the iteration screenshots. The hero section looks cool, but the rest of the front page could be refined even further (it still has an AI‑generated look).

When doing web designs with AI, I start with design tokens, typography, then build out every section by multiple iterations. Unfortunately, the more vague the prompt is, the more "AI‑generated" the end result looks.