Dialogue

In the early days of building Dialogue, I hired content creators to read books, extract insights, and together with AI, turn them into conversational podcasts. It “worked,” but it was a time sink. Every extra book meant more manual review, more hand-holding, more patch-ups. It wasn’t scalable, and my own goals started shifting from publishing more books to surviving the manual workload 🫣

That’s when I decided I've had too much 😆

Instead of trying to produce faster, I started automating the ugly parts.

➡️ First came book understanding and example-driven scaffolds.
Then podcast script creation.
But scripts kept showing the same issues. Updating the base prompt wasn’t enough, so I added a second layer:

➡️ Script improvement, fed with real examples of mistakes and how to fix them.
Still, things slipped through.

So I added
➡️ Script evaluation.

Then
➡️  Audio creation.
And of course—audio models make mistakes too. Listening to every episode was eating my life.

So I built:
➡️  Convert audio back to text → evaluate → compare against original script.

➡️ Next bottleneck: working with the content team. Spot checking, correcting, spot checking again. So I built a system where writers became fully self-sufficient, and every creator reviews another creator’s work.

And suddenly… the issues stopped.

I now listen to Dialogue the same way any user would. I don’t babysit the pipeline. I don’t chase edge cases. I don’t “check” anything unless I’m curious.

I’ve officially become a consumer of my own app.

And that one shift freed me up to focus on the business instead of fighting the product.

Dialogue turns books into conversational podcasts.