Vibecoding
Build for the vibe, debug later
Start new thread

What was the very first project you vibecoded with AI?

by
•

On Product Hunt, I can see many people launching their products using "vibe-coding tools" like , , or

I reckon many people who created something with them are usually developers who didn't have enough time for building a side idea before, but with AI, they could make it happen.

I am not very technical (know some coding/programming basics), but without the help of a tutorial or ChatGPT, I would hardly build a whole project.

Question not only for developers (but also tech newbies):

What was THE FIRST THING YOU VIBECODED?

  • Feel free to share the link or the picture

  • What tool did you use?

  • What was the most difficult part?

  • Did you earn any money with that?

Here is mine:
– It was supposed to be a directory of Bluesky tools

– I used by

– The most difficult parts were to define something + It also rewrote good parts of the code, so it was kind of a mess for me.

– I haven't earned any money because I haven't published the project. (I abandoned it. :D)

4.3K views

Add a comment

Replies

Best

I vibecoded on & . I think the most difficult part about this, was version control, just to add more context I'm a marketer by profession, and have 0 working knowledge about any of this until i actually began, so when I was halfway through it, things started breaking apart and I had no way to understand where the leaks where, it took me some time, some sessions with friends in tech to understand how these things work, and finally figure out a way to have more control over the code, even though it's vibecoded. I didn't earn anything out of it yet, but this did give me an idea for a bigger product with a more nuanced edge case. It was fun to build and see it develop into something people and the community genuinely talk about with interest.

 what if AI "damages" your code by constantly rewriting? Don't you need the help of a professional coder/programmer then?

 Absolutely, so far I didn't need that level of help (spray and pray would not work in the long term for sure), but I'm of the view that vibecoding is good for people like us, with 0 coding knowledge, or resource availabilty and drive to build something fast, I think this project helped me understand what to build next, find a PMF and stress test the idea. But when I actually build the next level of this, I'm not going to rely on this stack entirely, I may create a basic version to explain the idea to someone, or to visualize and plan how things will look like and work, but when I have to build a consumer focussed shippable version, I will almost certainly have to find a professional coder/tech resource to build it out with me.

I feel vibecoding is cool for low stakes, hobby projects, or to stress test an idea at a short time while you pool in resources and work on the plan for the bigger product, but when the stakes go high, and if you're planning on creating a data storage/get payments from the site, vibecoding can quickly become unreliable and expose a lot of cruicial security issues that might go very very wrong for high stake products.

My first "vibecoded" project is actually what I’m scaling right now:

I’m a founder with a product background, so I know the basics, but building a full-scale, production-ready tool was always a "someday" task. AI turned that "someday" into "this weekend."

  • What was it? A web-native hub to kill static PDF resumes. I wanted to create a standard where professional value is indexed via "Proof of Work" links (what we call the LCV standard).

  • What tool? Mostly used Claude 3.5 Sonnet paired with Cursor and a bit of Gemini 1.5 Pro for architectural logic.

  • Most difficult part? Implementing Privacy by Design. I didn't want a central user database. Getting the AI to help me build a "No-DB" architecture where all data is encoded in the URL state/local processing was a challenge, but it works perfectly now.

  • Did you earn money? Not yet in direct revenue, but the traction is wild: 200+ daily active builders organically. The real "money" right now is the signal we’re getting from the market.

It’s crazy to think that without "vibe-coding," this would still be just a 10-page PRD on my Google Drive.

 I would also recommend to add EN language. Or do you want to make it localised for Russia only?

My first real vibe-coded project wasn’t a toy app, it was shipping actual product work faster than I could before.

What I learned quickly: generating code is the easy part.
Keeping structure, handling edge cases, and getting to something actually reviewable is where things get real. That’s what made me much more interested in execution, not just generation.

My first vibe-coded project is actually what I'm launching tomorrow — OSS Discovery (), a directory of 1,284+ open source tools across 41 categories.

Used Antigravity + Claude + others heavily throughout. I'm a CS student who knows the fundamentals but wouldn't have shipped this solo in 2–3 weeks without AI assistance.

The hardest part wasn't the code — it was data. Scraped 7+ sources, deduplicated thousands of entries, cleaned messy metadata. No AI could do that part cleanly for me.

No money yet — it's free and launching tomorrow on Product Hunt Would love any support!

I vibe coded a lot but mostly for personal use web apps on my home server. In my experience, it is an advantage to vibe code if you have coding experience because you will be able to direct the Ai better and whatever terms and concepts the Ai throws at you, you will understand it easily. I do not use Ai wrappers, perhaps Ai wrappers are there for non-technical people. I use Ai directly to code. Using Ai wrappers diminishes your coding flexibility. I also do not use frameworks. Frameworks makes you less powerful and less flexible. Using Ai, I treat it as a speedup tool like giving me 10 sets of hands and brains. One piece of advice that saves you from bugs: NEVER ASK THE AI TO BUILD SOMETHING YOU HAVE NO SOLID UNDERSTANDING. If you ask it to build an accounting app but you do not understand accounting yourself that is a disaster waiting to happen.

Building out , i wanted to have an easy way to receive form submissions on for example my personal website which is pure html/js/scss. I didn't think at the beginning, but i rewrote the app 2-3 times and we now have 200 users up and running :)

I know this post is a bit old now but, I want to share my experience.

My first "vibecoded" project was a voice-activated expense tracker!

It uses audio recognition to listen to your voice and automatically logs/tracks your expenses.

It was mostly a fun experiment to see what's possible at the time, now I've completely changed how I think about building things!

 Is is online somewhere too?

 Now, I'm actually working on a far more complex project: a RAW photo editor that runs right in the browser. If my first app was a brick, this one is a skyscraper! It's already online and working, and I'm preparing for the launch soon.

 Wow, are you into photography too! :)

I had a technical background already, so I jumped straight into Cursor rather than using the no-code vibe-coding tools.

Later I switched to Claude Code, but the project eventually got finished and is actually being used today.

Like many side projects, it started from solving my own problem: organizing routines, checklists, and responsibilities for my kids.

The hardest part wasn’t writing code — it was defining the product, prioritizing features, and deciding what not to build. AI made implementation much faster, but product decisions were still very human 🙂

My first serious AI-assisted build has been a local-first code intelligence layer for coding agents.

The surprising part: the bottleneck was not generating code. It was giving the agent a reliable pre-edit model of the repo.

I ended up thinking about it less like “RAG over code” and more like a local static-analysis pipeline:

`repo -> parser -> symbol index -> reference graph -> impact analysis -> test discovery -> agent plan`

The core idea is that before an agent edits `validate_token()`, it should know:

- where the symbol is defined

- who calls it

- what imports depend on it

- what tests cover nearby behavior

- what files are structurally related

- what commands can verify the change

That feels very different from chunking files, embedding them, and hoping cosine similarity retrieves the right context.

I’m building this in the open here:

Still early, but it has convinced me that vibecoding needs a stronger context layer: understand → plan → edit → test → verify.

 will you launch it here?

 Thank you for the upvote and support, Nika! Yes, absolutely! It is still early. Right now, just focusing on gathering early feedback from developers using it with Cursor, Claude Code, and other MCP clients.

If you have a chance to try it out (via npm install -g open-kioku) or contribute, I'd love to know what you think!

hi and all,

its been a while since you asked that question. I feel like a lot has happened in the meantime. i do recommend claude code in vscode and the use of the AIWG framework. theres a whole SDLC as well as inception wizards etc. in there. there are others worth looking at but that is what I use.

i feel like we are just at the beginning of this. there will be models that spot security issues and fix them. we are at least moving from the vibe-coding term to an agentic engineering approch, dont you think?

it is all about being active and using these tools to create benefit.

check out some of my work at

best wishes

Tilman

 It doesn't mind that I asked before :) I am happy you shared your creation! :)

First
Previous
•••
91011
•••
Next