Most demos of AI apps look like the same demo: type prompt, get output, ship. Most production AI features look like that too — and you can feel it. They're fine. They're never great.

This talk is about what "great" actually takes. We'll build a URL-to-thread app live, but instead of writing one prompt and shipping it, we'll race three agents in parallel — each trying a different approach to the hook, the chunking, the voice — and merge the winners. Quality climbs as we go. You'll leave with a workflow you can use the next day on whatever AI app you're already building.

Built in Vite + React + Express. Racing on Ocean.

punchier, more directive. Loses the named insight ("first-take tax") which is the strongest pitch handle.