DjwalaAI — abandoned (honestly)

DjwalaAI was an AI DJ that built playlists from a chat interface — describe a vibe, get a playlist, click through to play in YouTube/Spotify. I shipped a working MVP. Got real users. Built a BYOK billing model on YouTube Data API. Set up OAuth with the YouTube scope I actually needed.

Then Spotify shipped the same feature natively, on their side, with their data, in their app. Same week.

I shut it down two weeks later.

What I got right

• Shipped the MVP in three weekends with a clean BYOK billing model that scaled to $0 marginal cost
• Auth + scopes + quota math were all clean
• Real users, real feedback within a week of launch

What I got wrong

• No moat. The minute a music platform decided to do this themselves, my product was a strictly worse version of theirs, with worse data and worse UX.
• I didn’t validate the moat before building. Building is fun. Validating-the-moat is not. Guess which one I prioritized.
• The lesson generalizes: any product whose only differentiator is “using AI on top of public APIs” is built on rented land. If the API provider can ship the same thing natively, they will, and you won’t see them coming.

What I do differently now

Before any new product, the very first artifact I produce is a one-page “who can kill this in one product release?” doc. If the answer is “the platform whose data I’m using,” I don’t build it. I’d rather spend the same weekend on something where the platform’s incentives are aligned with mine, not orthogonal.

This isn’t a sad story. It’s a $0 tuition payment. The actual cost was three weekends. The lesson cost $0 to absorb because I shut down fast. Founders who don’t shut down fast end up paying that tuition in years, not weekends.

The repo is still public if you want to read the code. The product is gone, but the architecture was clean.