There's no shortage of advice on how to vibe-code (or vibe engineer, if you're actually reading the code outputted by LLMs) better. We're inundated with how to write prompts, set up better context or skills for agents, or how to get the most out of different platforms or models. But talking to friends who've vibe-coded projects, the bottleneck doesn't seem to be getting started building, or that they can't get the agents to do what they want. We noticed two recurring patterns:

1) Most people never ship. Even though vibe-coding lowers the barrier to entry, time, and cost of building a MVP, it can also make it harder to ship when each additional feature or polish is just one prompt away. And sure, lots of projects are just scratching a personal itch, but if it solves your problem well, someone else out there would probably like it too—if nothing else, as an inspiration to create their own take on it.

2) Even those who do ship struggle to get anyone to try it. Your friends want to help, but they're often not the target audience. And as the cost of building decreases and the number of projects explodes, discovery only gets harder. Vibe-coded projects might have an even harder time than projects built by hand because audiences assume you didn't put as much thought or effort into them, or that they're not secure.

We used Lovable to build a StumbleUpon for vibe-coded projects. Visit the site, see a random project, try it out if it looks interesting, or click through to the next one until you find one that does look interesting. This probably isn't the most natural format for project discovery, but ranked leaderboards usually get gamified and feel like they have a higher bar for submission. Most of all, seeing what other people have built and shipped, especially when it's vibe-coded, makes shipping your own feel a lot more within reach. Try it out and let us know what you think!