The Limits Of Vibecoding

Yesterday I spent 3-4 hours “vibecoding” an app via Replit.

I was surprised how quickly I landed on a working prototype. With just a few prompts, after half an hour (most of that time I was waiting for Replit to finish a task), I ended up with a working app that allowed me to show items from a project management app on a calendar (including filtering on projects, teams, status, etc.)

It connected via API key to the project management app, all I did give as input that I wanted to have this calendar view and the name of the app. It looked up all information about the app and how to connect to it.

After a few hours, I was happy with the result and went to bed. Today, in the morning, the app no longer worked. After I tried to fix it via follow-up prompts and providing error messages, Replit stopped working altogether, and not only did the created app stop working, but Replit as a whole.

My takeaway is that “Vibecoding” is perfect for prototyping and creating internal apps. But you should be very careful about publishing something publicly, building apps, or even a startup product, with just prompting. After having some complexity, you can end up lost, and “vibecoding” can turn into “nightmarecoding”.

👀 The Creator's AI Handbook. Tools and strategies. Join 2,000+ AI-curious creators and get our FREE handbook.