I saw this on ycombinator. Someone has vibe-coded a recipe app.
I'm not even sure this isn't an April Fool.
There are some amazing recipes in there. Some of them might even be edible.
This is the future once our AI overlords take over.
EDIT: I think this one is actually real: https://www.recipeninja.ai/recipe/r_DdeXMdijiZW4QY/elon-musk-s-dirty-pants/show
I was trying to work out why ycombinator would want to get into that technical debt hell hole... then I realised its perfect for them they want to test a market and see if there are customers who will pay for this sort of stuff - then if there are turn it into a scalable, stable platform. Interesting name choice, I imagine the Ninja cookware trademark lawyers are currently earning their fee! But again that might not be catastrophic for a seed funding incubator - a sneaky move to get some extra publicity?
I was trying to work out why ycombinator would want to get into that technical debt hell hole... then I realised its perfect for them they want to test a market and see if there are customers who will pay for this sort of stuff - then if there are turn it into a scalable, stable platform. Interesting name choice, I imagine the Ninja cookware trademark lawyers are currently earning their fee! But again that might not be catastrophic for a seed funding incubator - a sneaky move to get some extra publicity?
A New Zealand supermarket experimenting with using AI to generate meal plans has seen its app produce some unusual dishes – recommending customers recipes for deadly chlorine gas, “poison bread sandwiches” and mosquito-repellent roast potatoes.
https://www.theguardian.com/world/2023/aug/10/pak-n-save-savey-meal-bot-ai-app-malfunction-recipes
I for one welcome our new robot overlords and their recipes for turpentine French toast and bleach infused rice.
It's worth reading this... https://www.theguardian.com/technology/2025/mar/16/ai-software-coding-programmer-expertise-jobs-threat
The app likely already existed, and the LLM has copied and modified its codebase, which is clever, but it would not be possible to code something like this up using an LLM unless there was a very similar starting point.
The chocolate and sardine tart looks exquishit.
I was trying to work out why ycombinator would want to get into that technical debt hell hole
Its on their hacker news forum vs an actual ycombinator funded firm from what I can see. It does seem to be the dream combination of AI code and AI content. What could possibly go wrong.
I was trying to work out why ycombinator would want to get into that technical debt hell hole
Its on their hacker news forum vs an actual ycombinator funded firm from what I can see. It does seem to be the dream combination of AI code and AI content. What could possibly go wrong.
You're 100% right there. It's just on Hacker News and not a Y Combinator funded company.
It's worth reading this... https://www.theguardian.com/technology/2025/mar/16/ai-software-coding-programmer-expertise-jobs-threat
The app likely already existed, and the LLM has copied and modified its codebase, which is clever, but it would not be possible to code something like this up using an LLM unless there was a very similar starting point.
It probably did exist, because it's a reasonable idea for an app, but that's not really how this works. These tools will make any nonsense app you want them to, whether anything similar existed before or not.
It's a great tool for programmers to aid productivity. It gives you a starting point and sometimes ideas you'd not have thought of. But non-coders generally just create garbage with it. It might look like it's working but I guarantee any programmer will find major problems with just a few minutes looking at the code.
Real example: user wanted to add LDAP (basically log into an intranet webapp using your windows login) functionality to an existing legacy app. They used ChatGPT etc to write a replacement login function. It had some problems so they asked my company to look at it.
When I looked at it in the debugger, the code contained a call to a function with more parameters than the language actually had defined for that function. Turns out ChatGPT had simply tapped the code out as if that feature existed. Except it didn't. I guess it had been trained on a page that said "it would be nice if function abc(x,y) also had parameter z like abc(x,y,z) so I could do <thing>".
Of course you'd think that would throw an error, but they'd also messed with the error handling so it was now silently ignoring all the notices and warnings the code was generating.
End result: the login code worked for exactly one department, everybody else could simply click "login" and bypass the login process entirely!
