I've just 'vibe coded' an entire app in Cursor – check it out: FeedBagel.com.

It looks fine and works great, but the whole process left me wondering if I should ever have bothered. I managed to ship this MVP, but I sometimes find the frustration of vibe coding (in its current state) worst than just coding myself.

Could I have done this faster by thinking?

Doubts aside, in the spirit of this AI experiment, I'm persisting to see where it goes. To take it further, I've handed the role of product manager to AI too!

Perplexity: AI as Product Manager

FeedBagel started out as an internal tool within Letter that fetches article feeds from websites for newsletters, but I thought this can be useful to anyone – the perfect excuse to let Cursor run wild.

After getting the first version working quite fast, I wondered where this project might go – is it an app that I'll just be using, or could it be its own business?
Why not let AI determine its fate?

To do so, I set up a Space within Perplexity that'll hold all the project's information, and play the role of Product Manager:

I've filled the space with references and relevant links, and even generated details of the current tech stack and features using Cursor. That gives it the knowledge base from which to decide the roadmap:

The summary table of the Early Growth Plan looks like this, and I'll from here just tell Cursor to do what Perplexity advises:

If you want to see more, I posted the process on YouTube:

Plus, I've been experimenting with some AI-generated voiceovers and AI Avatars. These could help improve little screen recordings if used well. Here was the first version (it's a bit of a joke with the burger guy):