Eurovision Scrollytelling
I made this!
I've done a data analysis of Eurovision lyrics until 2025, and Claude Coded over 8 weekends to create this scrollytelling article. Enjoy!
(, Fri 8 May 2026, 12:34, Reply)
Claude coded = vibe coded with Claude?
Just trying to keep up with AI slop lingo
(
simbosan is too tired, Fri 8 May 2026, 13:23,
Reply)
Not quite. And if you look my previous work, I've done data analysis for about 20 years.
The toolkit of today includes LLMs, but using Claude Code and vibe-coding is not the same thing. This was the result of about 8 weeks of back and forth coding, testing, quality checking, etc.
Wait for the blog post for the how to and the jupyter notebook, but the slagging off I'm not finding fair without checking your assumptions first.
(, Sun 10 May 2026, 13:44, Reply)
Also, this is from about 8 years ago: brexit.puntofisso.net/
Same tech (including the same Scrollama library), same animation style etc. The fact professional developers use LLMs today to help their coding is not what "vibe coding" is.
If you think this is "AI Slop" I'm sorry but I don't agree with your definition o "AI Slop" (cfr en.wikipedia.org/wiki/AI_slop).
If you are technical, I'd recommend Simon Willison's blog where such topics are extensively discussed in rather intellectual terms.
(, Sun 10 May 2026, 14:04, Reply)
Read my comments to other messages. "Fuck off" is an odd and rather unwarranted comment. Happy to answer questions, though.
(, Sun 10 May 2026, 13:45, Reply)
It's not a very odd or unwarranted comment around these parts.
The oddest thing about it is that I neglected to also call you a cunt.
Think of b3ta as the Mos Eisley cantina, and think of AI as your droids. You need to leave them outside, they're not welcome here.
(
brb Monkhouse Stampede, Sun 10 May 2026, 16:51,
Reply)
Is the content and data analysis AI?
I suspect saying it is Claude Coded has caused an AI;DR reaction to this, which may be undeserved.
(Though informing Richard Dawkins' girlfriend about accessible colour schemes may be required - we are old and some of that is a bit unreadable)
(
monkeon schmonkeon, Fri 8 May 2026, 15:31,
Reply)
Thank you for saying that. I've been at this for a long time so I'm not surprised. I'll blog about the experience. But I think there's a fundamental misunderstanding about equating vibe-coding (Lovable etc) with using LLM tools to help one code.
(, Sun 10 May 2026, 13:47, Reply)
Fair re colour scheme, though. I tried to use mutually exclusive colours but it's hard to do a fully accessible story of this kind. I've been also thinking about using symbols rather than colours but that would kill the performance even more strongly (1795 SVG dots aren't easy to move around, especially as the grouping is based on force field).
(, Sun 10 May 2026, 14:01, Reply)
To address your question, the workflow that I used the LLM for included:
1. creating a web tool to manually inspect and correct the lyrics (originally coming from a JSON file); the LLM helped here with creating the web page, saving back to the file, etc
2. creating and evolving a number of jupyter notebooks to do specific bits of data analysis; the LLM here helped with the coding of automating the workflow, but each bit of analysis was co-created (and checked); if anything, the LLM helped brainstorm (e.g. "for this amount of data, would LDA work better than BERTopic analysis and why")
3. iterating quickly on the D3 basic animation I had from a previous project; things like "try this type of transition" are very effective
All in all, I'd say again that "Is the content and data analysis AI?" is not a question that can be answered in a yes/no way. Not for this project, nor for any professional developer work today. The data analysis was created by a "conversation" that evolved over about 2-3 months and that required interruptions, code-checks, code rewrites (either by me directly or by instructing Claude to do changes).
Many professional developers use LLM tools as they make our work faster and more effective. I constrain myself to using LLMs only with libraries/languages that i know and understand well, and that I can fact-check and quality-check.
If by "content" you mean the text, it's not "AI generated". I had Claude create placeholders in the Scrollama "cards", which I then wrote by hand. I'd never use a LLM to write text (I've been sending a newsletter about data for 12 years and it's all written by me) because it would lose my "voice".
I hope that makes sense, but if you have any questions I'd be happy to answer. I'm drafting a blog post about the experience of coding using Claude for this, and any point I should develop would be much welcome.
(, Sun 10 May 2026, 14:14, Reply)
Fair enough, but I'd question the faster and more effective for professional development bit. Most vibe coded sites I've ever looked at seem to be by people wanting to call themselves full stack devs without enough understanding of the frontend developer, UX designer, technical SEO etc jobs to spot the glaring mistakes (and vice versa, hence all the hackable sites built by people without any server-side or security experience). When I experimented with it, it was churning out code that's been deprecated since 2008, and you had to instruct it not to, for example. I know boosters would do a mea culpa about prompting it wrong, but if it's default output is broken, that's bad software.
My experience with it has been that beyond some code review and writing individual functions, I really do not trust it's output without full review (which soon eats into the claim that it is faster). Also, having to write an essay to generate something professionals could write quicker in code doesn't actually seem very time saving, especially when you know that there will likely be 10% or so that needs fixing after it has been generated because your prompt left some assumptions open and it chose a mad solution. Most people already had their own libraries for the repetitive work, and there were plenty of generators and libraries like Google Charts that churned out consistent code for displaying data.
I see why it works for prototypes and personal projects (which I yours falls under) - until it becomes prohibitively expensive, at least - but all the examples I've seen of fully vibe coded sites seem buggy and destined for major future problems.
I am aware the above is a general moan about vibe coding than related to your actual site, and a bit of a wall of text, but it sometimes feels like I must be missing something, yet the more I look at Claude Code, the worse it gets. All the people doing videos on it seem like slick influencer salesmen rather than the awkward nerds doing a deep dive that I'm used to.
(
monkeon schmonkeon, Sun 10 May 2026, 15:34,
Reply)
I agree with you re vibe coding websites (like Lovable?). But that's not a tool I'd ever use, especially for a data analysis. Using Claude Code in a development environment is not quite the same thing, though, especially in terms of structuring a project (for example, my project init starts always with the library and version I want to use, e.g. D3 v2.7).
The speed comes from the writing coding within guidelines/guardrails. But it's then down to the developer to check those (e.g. unit testing etc). I don't think it's, overall, a 100x in speed, but it's well close to a 10x. For comparison, the Brexit vote explorer I linked above took me about 4 days to code, while this analysis has taken me over 2 months – the complexity was not the actual code.
Overall I agree with the sentiment on "vibe coding", though. It's just that I'd distinguish the "hey create a website that does this" (which I abhor) from the LLM-aided coder (which I fall under).
(, Sun 10 May 2026, 15:59, Reply)
Full stack dev here, hello (many, many years experience)
Big difference between "vibe coding" and "AI assisted development". Scaffolding shit (that can be done with my eyes closed) but took a good chunk of time to do now takes minutes. Good devs are constantly keeping it in check, and there's a fair amount of ballsups once things get a little complex. It pretty quickly fails to follow patterns I've found and just starts doing its own thing (and the annoying loops it finds itself in). I've not trusted it with anything too sexy, but it definitely speeds up the donkey work. Fuck the 50 million "Hey I coded a tailwind component library" bros though. There's going to be a wealth of oversaturated, useless, cookie-cutter shit over the next few years.
(
Fadgebadger Sweaty sumo wanks for everyone!, Mon 11 May 2026, 20:21,
Reply)
Are people actually getting financially rewarded
for doing more high-end code and less donkey work, do you know? Or are they just expected to do more burn-out work for the same money?
Are clients charged the same, or have you now got to get more work in to make the same?
www.youtube.com/watch?v=hfionMmCGoU
(
monkeon schmonkeon, Tue 12 May 2026, 14:59,
Reply)
Maybe @Fadgebadger can reply in more informed ways than I can, but this probably depends on what sort of contract you have.
In corporates, "how much" you work is fixed*. If you're a freelancer, there's obviously two ways this can go (are you paid by project vs by the hour/day etc).
(*) obviously there's that philosophical question about "should we all work less thanks to automation", but that's beyond this discussion
(, Wed 13 May 2026, 8:52, Reply)
But is there any evidence
that a dev's pay has actually increased since LLMs arrived and their "productivity improved". No one positive about AI ever seems to talk the economics.
My assumption is that either clients benefit at the cost of the business, as they are told by boosters that coding is now a piece of piss and don't understand that something they can generate which looks 70% there is very different to being 70% there, and so they believe the robot will do most of the work (then it becomes a race to the bottom as code is expected to be cheap and devs/sales teams burn out as they now have a far harder job to do), or the company benefits and keeps their prices the same (at the risk of cheaper rivals undercutting them).
High-end dev work *should* be paid more, as that's harder and burns you out faster (arguably a lot of the work AI is doing is the work you can still do when your brain is done for the day). If people are doing more of the hard stuff but aren't paid more-per-hour then they're being screwed over.
Freelancers have to do both sales and dev work, so doubly exhausting.
I don't see how AI can possibly do anything but destroy what was a good ecosystem. We used to share knowledge with each other, now just give it to the billionaire-owned robots which destroyed those communities by stealing from them.
People who think they won't be replaced if it eventually does what the AI companies promise aren't considering why their bosses are so keen on them documenting all their thought processes in little .md documents, which the company own.
(
monkeon schmonkeon, Wed 13 May 2026, 10:35,
Reply)
I can speak for my company, but that is obviously not every one
We're a smallish dev house, we're big on wellbeing. I give my devs an amount of work in a sprint, if it's done, it's done it doesn't matter what it took to get to that point. If we use a tool to make us more productive then my two cents is that improvement is given back to the team in lower project pressure and better work-life balance. Our projects are generally costed based on complexity and skill, not necessarily timescales (though they are a factor, albeit a smaller one). We've also increased pay again this year for our devs. AI dev has a cost in itself of course, but the ROI is good enough to keep it around. We've not changed our general project costs we charge to clients, but we've mostly kept them as low as possible for years now. I have no inclination to replace my good developers with AI, but I'm certainly not going to pass up a tool that makes them happier and more productive.
Edit: probably noting that I totally see your (and most others) point about AI, and just because I see it this way, doesn't mean your points are not valid or true.
(
Fadgebadger Sweaty sumo wanks for everyone!, Wed 13 May 2026, 11:50,
Reply)
Thanks.
To be honest, I don't think my use is that different to yours at the moment. I'm just probably more focussed on the negatives, and can't see any positive way it could play out long-term for the majority of developers unless things freeze at this point or go tits up for the AI companies (which isn't an impossibility).
(
monkeon schmonkeon, Wed 13 May 2026, 12:18,
Reply)