About this session

Every SaaS CEO has an AI strategy, fewer have sat down, written a prompt, and built something that ended up on the product roadmap. Klaviyo's CEO built the company's latest marketing agent, called Composer, from a single prompt to production.This session is a candid walkthrough of what actually happened: the first prompt, the architecture decisions, where the agents got stuck (and where they didn't), how you know you're on to something and what it took to go from prototype to product. The deeper lesson isn't about code — it's about what separates AI that works in the real world from AI that dies in demo. Agents aren't smart enough on their own. They need the data layer, the identity layer, the customer context and the infrastructure that most companies don't have. This is a session about what it takes to build software in the autonomous era — not theoretically, but from someone who got their hands dirty. What we'll cover: •The original prompt Andrew used to kick off the build — and how the requirements evolved from there •The specific moment the prototype became something worth shipping — and what the path from proof-of-concept to production really cost •Why AI-built software lives or dies on the quality of data underneath it, and what that means for how B2C companies should be thinking about their infrastructure •How this experiment changed Klaviyo's product philosophy — and what it signals about how software teams will operate in the next few years