Play with it » Watch the video »
I’ve just made the most ambitious website of my career, and AI wrote about 95% of it.
Writing that down makes me a bit uncomfortable. I’ve been making websites since I was 15 and doing it for a living for over 20 years. Two years ago I wouldn’t have believed this was possible, and I definitely wouldn’t have believed AI could help make something I’d call creative.
If I’d just asked it for a website, I’d have got complete shit. Four times, as it turns out.
The web used to be weird
When I started, there weren’t really any rules. No cookie banners, no GDPR, no playbook for what a website was meant to look like. You typed in an address and had no idea what was about to happen.
Flash made it a playground. Websites could be games, toys, experiments, a load of noise for no reason. The Space Jam site from 1996 is still online, still a starfield with planets for navigation, and still more fun than most of what I’ve seen this year. StumbleUpon was magic because every click took you somewhere stupid.
Then everything got better. The iPhone, the mobile web, Flash dying, responsive design, accessibility, web standards. Websites now work on everything. You could probably open one on a fridge.
All of that is good and I’d never go back. Somewhere along the way, though, we made the web work for everyone and made a lot of it really boring.
The challenge
Genius Division already had a perfectly decent website. I didn’t need another one. I wanted the opposite: something strange and tactile that made the internet feel unpredictable again, and that nobody could ever call a template.
There was a second experiment running alongside it. I didn’t believe AI could be creative, so I told Claude that. I said I thought we’d end up with another generic corporate website and asked it to prove me wrong. I’d work as hard as it did, and we weren’t stopping at competent. We were stopping when something surprised me.
Of course AI can build a website. I’ve built one in 20 minutes just to see. What I wanted to know was whether it could help me make something properly creative.
Claude proved me right (at first)
The first proper attempt was a website that looked like a computer desktop. Windows you drag about, icons down the side, a little lock screen on your phone. Claude was clearly pleased with itself.
It was fine. Slightly twee. And we’ve been making websites like that since the Flash days (Windows 93 took the idea much further and is far funnier).
The next one was a corporate website where you drag a handle across the page and it turns into a tired 2012 template, with a “Best viewed in Internet Explorer 8” badge. Actually quite good. Still an idea I’d seen before.
After that it was mostly me saying no. Too generic. I’ve seen this. Push it. More colour looked childish, and less colour looked smarter and even more forgettable. At 5:39pm on the Sunday I typed “I still feel like we’re too generic and not getting the depth of the feel I want”, and seven minutes later I stopped the whole thing.
AI can generate endless options. Somebody still has to know which ones are shit.
Starting from taste
So I started again in a fresh chat, and this time we didn’t start with a website at all. We started with a mood board of 50 things from every era of design: Victorian playbills, Bauhaus, Swiss posters, Braun radios, British road signs, punk zines, rave flyers, Space Jam, jodi.org, Bruno Simon’s portfolio where you drive a little toy car around. Felix Rieseberg’s website was another one that set me off.
I marked every one, and the pattern was a bit embarrassing. In print I like strict, tidy modernism, grids and systems. Online, every site I loved was a bit mad and made by someone who hadn’t asked permission.
The thing in the middle turned out to be The Designers Republic, the Sheffield studio famous in the 90s for Warp Records sleeves and the WipEout games. Swiss grids and corporate pictograms turned into something funny and acid. Just down the M1 from our office.
When I was asked what I wanted, I said “a website like TDR should have built but never did. A Flash website but it’s not a Flash website, made usable.”
Nearly
The first go at that was built around British road signs, with a motorway sign for a menu. It made me cringe. “Too safe,” I said. “Show me the extreme of this.”
What came back was colourful and strange and nowhere near what was in my head. The top of the homepage was covered in fake adverts. A hazard label saying “Caution: may cause growth”. A voucher “good for one (1) honest opinion”. A No Strategy sign, because I’ve banned the word in our office.
That was the first moment it felt like a collaboration. It looked good and it worked, and most people would have stopped there.
I nearly did. I told it “this is not my floor” and asked for something even more extreme, mostly for a laugh.
Oh shit
What came back wasn’t really a website any more.
The ads had real physics. They rained down from the top of the screen and you could pick them up and throw them. Things made noises. There was music. There was a button saying “Don’t press this” that smashed the headline into loose letters you could chuck about. There were secrets, and stupid little interactions that had no reason to exist.
It felt like Flash again, built with today’s tools, and it still works with JavaScript switched off and loads fast on a rubbish phone. We roast other people’s websites for being slow, so it had better.
I didn’t know if I’d pushed it too far, so I sent it round. The team, mates, my other half. Everyone I showed it to started throwing the ads about. Our office manager couldn’t stop playing with it.
That was version 0.1. It’s grown a lot since. There’s a vending machine (B4 is sold out). There’s a fruit machine that tells you what your website costs. There’s a shelf of VHS tapes you can stick in the telly.
There’s a trophy cabinet with things in it you’ll probably have missed. And for reasons I can’t fully explain, there’s a Flappy Bird style game hidden in there somewhere. I’m not telling you where.
So did I design it?
This is the bit that makes me uncomfortable.
It’s the most creative website I’ve made, and I didn’t directly design or build most of it. AI wrote the code. It did a lot of the visual work. It suggested ideas I’d never have had.
So what did I do?
I knew when something wasn’t good enough, and I said no to hundreds of decisions. I brought the references and set the direction, then changed it. I knew what had been done before, and I knew when Claude thought it was being original and wasn’t. I kept pushing when a perfectly acceptable result was already sat there.
After all these years, the most valuable thing I’ve got might be knowing when something is shit. AI makes that worth more, because now anyone can get to “fine” in about four minutes.
So, is AI creative?
It can obviously build a website, and it can knock out competent, generic work really quickly. Whether a person and AI can make something creative together is the question I actually cared about.
This project changed my answer to yes, with a catch. The first answer wasn’t creative, and neither was the second, the third or the fourth. The creative bit came from the back and forth: it generates, I reject, it tries again, I bring something it hasn’t seen, and round we go until something makes me swear at my laptop in a good way.
We’re using it wrong
Most AI is sold as an efficiency machine. Write this faster, code this faster, save 30 minutes, make more stuff.
That’s fine, but it mostly gets us more of the stuff we were already making, just quicker. There’s plenty of that already.
The question I care about now is what I can make that I couldn’t make before. Two years ago I couldn’t have built this website. Now I can, and the next one can go further.
So next time you open Claude, or Codex, or whatever you use, don’t ask it for the thing you’d have made anyway. Reject things. Make something stupid and unnecessary.
Then go and throw some ads about.