What are you vibecoding currently?

SpeedBus

After looking at the number of tabbed SSH clients/termius clones that people vibe coded I decided to vibe code one to see how far I could get, surprisingly claude fable was able to build a decent enough looking app and also write tests, flag certain gaps in the way the app security was etc which is kinda cool to see, that got me thinking what other interesting/cool projects others are working on?

so.. share your vibe coded projects below! Screenshots or project links!

Advin

We've recently been working on a new homepage/website with assistance from certain LLMs. I take issue with frontend UIs that are generated from LLMs with minimal tweaking (it's usually very easy to typically tell when a UI design is generated with AI), but it has been an immensely useful tool for expanding upon a specific design given to it and iterating upon it quickly.

Another project idea I had, which can probably use some help from AI assistants, is an internal asset management system. Perfect project to use LLMs on since we just need something quickly. I think it would be cool to integrate support for shipment tracking numbers to track inbound components and have it be able to print asset tags for our hardware to keep track of everything.

mikho

https://lowendspirit.com/discussion/11268/vibe-coding-a-vestacp-alternative#latest

Web hosting panel.
Open source for everyone to check out.

yoursunny

We wanna vibe a script to verify a routed IPv6 subnet is actually routed.
The VPS streams its tcpdump of ICMPv6 traffic to our validation service, while the validation service pings random addresses within the VPS subnet.
Based on whether the tcpdump sees Neighbor Solicitation or echo requests, the validation service can assess whether a routed subnet is real or the provider has mis-represented their service.

AnthonySmith

an AppImage port of "The Adventures of Robin Hood" 1991 computer game originally released on the Amiga.

poei

Bierovic Momentum Retracement Trading Bot

Faithful Python implementation of Tom Bierovic’s Momentum Retracement method (from 18 Trading Champions, 1996).

kf5ydu

A website to host my parents wedding video ☺️

rcy026

I know this is a unusual stance but I am a lousy programmer and I am well aware of that, so I do not consider myself competent enough to use AI to code. A lot of people are like "I cant code so I use AI". In my mind, that is like saying "I do not have a driver license so I drive a car", or in some cases even "I am extremely clumsy and legally blind so I will now attempt to drive this excavator trough this public mall".

I do however use AI, just not for code. It's awesome for graphics work. "Hey Claude, get me a theme for this, use these colors" or "Make some graphics for this presentation". I'm about as good with graphics as I am with code, the difference is that my competence in judging whether a picture is good enough to use or not will never result in a security issue. It might result in a HR issue (yeah, that happened), but never security.

I also use AI heavily for one time internal work. Move all data from this ipam to this new ipam, convert this config, get data from this api and convert it to this, etc etc. One job, run the code once, never use it again. It saves me enormous amounts of time and the risk of anyone being able to exploit it is very close to zero.

tbnuser

Couple of things:

1. counterfactual analysis on my admittedly very stupid IPO purchases and comparing them to a 70/30 split of two index funds to see if I'm better off or not - this was purely vibe coded with requirements documents, specs and steering. no link yet since I'm still refining what a "v1" would be
2. going through my "old" golang codebases and seeing what changes should occur, and using Opus / Gemini 3.8 Flash to "teach" me what to do and what to write
3. updating my personal site blogs. I write a .md file, structure it with images, and get an LLM to convert it to pure html+css based off my existing (hand written) blog pages. Saves me copying and pasting boilerplate html and results in an ultra low size website

Neoon

I took a harness, let the LLM modify it, and gave it a task to code me a "airbnb cli script to list the cheapest prices"
It nailed it, now I added booking.com support.

motafokabr

Using AI assisted development (for the frontend mostly) to get a browser extension for ProxiedMail.

The API integration I managed on my own, I work as a backend developer, but my frontend and design sucks....

poei

@rcy026 said: In my mind, that is like saying "I do not have a driver license so I drive a car"

I look at it more like "I do not have a driver's license, so I pay someone to drive me around."

@rcy026 said: I'm about as good with graphics as I am with code, the difference is that my competence in judging whether a picture is good enough to use or not will never result in a security issue.

In my projects use case, that is what backtesting is for. I would never use any financial software instrument, such as a Forex EA for example, without thorough backtesting of it with historical data first.

@rcy026 said: Move all data from this ipam to this new ipam, convert this config, get data from this api and convert it to this, etc etc.

So you're saying absolutely nothing could possibly go wrong, or mistakes happen with using AI for vibe coding this kind of stuff? For example say you ask it to write a script using rclone backend to move all your data between locations. How do you know it's not going to add an incorrect flag or something and ruin your data?

I suppose for that example is why the --dry-run flag exists in rclone, but that's exactly what I was talking about with backtesting. Anyone that uses AI coding, and then doesn't do any modicum of isolated testing before throwing it in to production, deserves what they get. You don't need a computer science degree to do that.

MannDude

Even if you don't use AI to code, you can use it to learn to code. No more copy/pasting an error log entry into your search engine of choice, reading some reddit thread from 6 years ago and them some Stackoverflow submission from 12 years ago to try to figure out something obscure.

rcy026

@poei said:

@rcy026 said: Move all data from this ipam to this new ipam, convert this config, get data from this api and convert it to this, etc etc.

So you're saying absolutely nothing could possibly go wrong, or mistakes happen with using AI for vibe coding this kind of stuff?

No, that is absolutely not what I'm saying.

For example say you ask it to write a script using rclone backend to move all your data between locations. How do you know it's not going to add an incorrect flag or something and ruin your data?

I don't. I get the impression you are trying to make a point here but I fail to see it.

I suppose for that example is why the --dry-run flag exists in rclone, but that's exactly what I was talking about with backtesting. Anyone that uses AI coding, and then doesn't do any modicum of isolated testing before throwing it in to production, deserves what they get. You don't need a computer science degree to do that.

I guess it depends on your definition of "testing".
I could ask AI to build a complete hosting solution with payment portal and whatever and test it for weeks or even months to make sure everything works. But, even if it passes those tests with flying colors and every function works perfectly, it does not in any way mean that it is secure! The code quality may be terrible and it might be riddled with bugs. I'm not saying you need a computer science degree to know that, but you do need to be a good coder!
Just look at the number of various panels that people have proudly presented here. Many of them have spent a lot of money on tokens and most likely hundreds of hours of testing before proudly showing their project to the world, and the people here on LES have found enormous security issues and exploited it literally within minutes.

Neoon

VibeCoding on all 3x KS-LE-B:
https://stats.neoon.net/dedi
Let her RIP BABY.

So far results better than expected.

poei

@rcy026 said: Just look at the number of various panels that people have proudly presented here. Many of them have spent a lot of money on tokens and most likely hundreds of hours of testing before proudly showing their project to the world, and the people here on LES have found enormous security issues and exploited it literally within minutes.

In the same token, look at all the vulnerabilities and flaws that LLM such as Claude 4.6 has found in popular software that "great coders" had been auditing and maintaining for years. It works both ways.

@rcy026 said: I don't. I get the impression you are trying to make a point here but I fail to see it.

You also said this..

I also use AI heavily for one time internal work. Move all data from this ipam to this new ipam, convert this config, get data from this api and convert it to this, etc etc. One job, run the code once, never use it again. It saves me enormous amounts of time and the risk of anyone being able to exploit it is very close to zero.

Sure, there is little to no chance of anyone else exploiting that code locally, but there still is some chance that the code you run once is flawed or buggy. All it takes is one error or overlooked flag or what not, one time, and your data is corrupted. You're trying to give the impression that using AI is ok for this one time job, but not for other software or scripts/jobs that are used more than once, in the wild. The answer is, both carry inherent risks. I'm not saying all vibe coded stuff is awful, but you're attitude towards this seems a bit wishy-washy, if that is the correct term.

rcy026

@poei said:

@rcy026 said: Just look at the number of various panels that people have proudly presented here. Many of them have spent a lot of money on tokens and most likely hundreds of hours of testing before proudly showing their project to the world, and the people here on LES have found enormous security issues and exploited it literally within minutes.

In the same token, look at all the vulnerabilities and flaws that LLM such as Claude 4.6 has found in popular software that "great coders" had been auditing and maintaining for years. It works both ways.

@rcy026 said: I don't. I get the impression you are trying to make a point here but I fail to see it.

You also said this..

I also use AI heavily for one time internal work. Move all data from this ipam to this new ipam, convert this config, get data from this api and convert it to this, etc etc. One job, run the code once, never use it again. It saves me enormous amounts of time and the risk of anyone being able to exploit it is very close to zero.

Sure, there is little to no chance of anyone else exploiting that code locally, but there still is some chance that the code you run once is flawed or buggy. All it takes is one error or overlooked flag or what not, one time, and your data is corrupted. You're trying to give the impression that using AI is ok for this one time job, but not for other software or scripts/jobs that are used more than once, in the wild. The answer is, both carry inherent risks. I'm not saying all vibe coded stuff is awful, but you're attitude towards this seems a bit wishy-washy, if that is the correct term.

The difference is that if the code that I run (locally, once, for myself) is buggy, the most damage it can do is possibly destroy my data. Data that I of course have backed up like a million times before I attempt to migrate it anywhere, hell, in every situation I can think of I would run the code on a copy of the data first just to make sure it works. To put it simply, even if the absolute worst possible scenario were to happen, it still would be about as serious as spilling a cup of coffee. Just clean up the mess and try again.
If you think this is even remotely similar to using AI to code entire panels, portal, websites or software that is published on the internet and handling stuff like personal information, logins, even payment and credit cards, then I sincerely hope you are not involved with anything that has to do with IT security. Do you seriously not see the difference between using AI for one time, one purpose personal scripts versus building entire portals or security software for public and often even commercial use?

MGarbis

My childs wants to learn to type quickly and without errors, so I created a game and practice page for them. The end result is a single HTML file that could be used online, but it's currently being used locally. It includes two different practice modes. I am surprised how good job AI made.

poei

@rcy026 said: the most damage it can do is possibly destroy my data

Well yeah, that's not so good is it?

@rcy026 said: Data that I of course have backed up like a million times

@rcy026 said: If you think this is even remotely similar to using AI to code entire panels, portal, websites or software that is published on the internet and handling stuff like personal information, logins, even payment and credit cards, then I sincerely hope you are not involved with anything that has to do with IT security. Do you seriously not see the difference between using AI for one time, one purpose personal scripts versus building entire portals or security software for public and often even commercial use?

Don't get all wound up in a knot about this. I just think you're having some comprehension issues here. Read again what I posted, and then point out to me what you don't understand about it. Specifically, point out where I said that public facing vibe coded software carries no risks.

@poei said: The answer is, both carry inherent risks.

What I actually said here is that was BOTH public and private use carries some form of risk. Not EQUAL risk, but still risk. If anything, the public code (ie. a public vibe coded Github repo) could still be audited by other people. The private code you use for yourself is only audited/tested by yourself. Would I use a vibe coded billing panel for my hosting service (If I had one), probably NO. Would I use a vibe coded python script to migrate data to/between my B2 and S3 and storage boxes, YES, and I would raw dog it without backups, because I like living on the edge.

Do I work in IT? No. My educational background is in Finance. However, I have been using computers in some shape or form since around 1982, a little bit before "IT" was even a thing.

Do I need a diploma from DeVry to understand risk? I don't think so.

Do I understand risk, and levels of risk? You betcha.

I really don't know what else to say about this lul.

mikho

@rcy026 said: Data that I of course have backed up like a million times before I attempt to migrate it anywhere,

rcy026

@poei said:
What I actually said here is that was BOTH public and private use carries some form of risk. Not EQUAL risk, but still risk. If anything, the public code (ie. a public vibe coded Github repo) could still be audited by other people. The private code you use for yourself is only audited/tested by yourself. Would I use a vibe coded billing panel for my hosting service (If I had one), probably NO. Would I use a vibe coded python script to migrate data to/between my B2 and S3 and storage boxes, YES, and I would raw dog it without backups, because I like living on the edge.

So we are basically saying exactly the same thing. Using AI for personal, local one time jobs is fine, but using it for production code with high demand on security is a big no no unless you really know what you are doing. Pretty much exactly what I said but you questioned. So what are we arguing about?

Do I work in IT? No. My educational background is in Finance. However, I have been using computers in some shape or form since around 1982, a little bit before "IT" was even a thing.

That makes two of us. The difference would be that I do work in IT and has been doing so since the 80's.

SpeedBus

@SpeedBus said: tabbed SSH clients/termius clones that people vibe coded I decided to vibe code one to see how far I could get

FWIW -- with Opus 5.5 now, it can build the app, and even take screenshots with dummy hosts in it / features demo'd in it.. seems to somehow be on par with Fable 5.1 it seems like~