Autonomous Marketing at Klaviyo: Architecting Composer and 1:1 Personalization at Scale

This talk covers the our vision of autonomous marketing at Klaviyo, and goes inside the engineering behind Klaviyo's autonomous marketing system — the architecture, tradeoffs, and infrastructure that power AI-driven customer experiences across nearly 200,000 brands, 7.3B+ consumer profiles, and more than 2 billion events processed every day.

The centerpiece is Composer, Klaviyo's AI agent for autonomous marketing. Composer takes a marketer's idea to a launch-ready campaign in minutes, and continuously audits marketing performance and generates account-specific recommendations. We'll walk through the agent architecture behind this, lessons learned and how we maintain quality and brand consistency at scale. 

Running alongside Composer is a 1:1 personalization layer that determines the right message, audience, time and channel to communicate for each individual customer in real time. We'll cover the ML models powering these, and how they're trained and updated continuously across a multi-tenant system. Both are grounded in the Klaviyo Data Platform (KDP) — processing over 1.6 billion API calls on an average day and serving as the real-time data foundation for every agent decision and personalization signal. 

We'll discuss the architecture choices that make this work at our scale, and where the hard problems still are. We'll also cover how we're building interoperability through our open MCP Server infrastructure, and the interesting challenges this introduces as agents start to operate across the broader AI stack.

This is a look at what it actually takes to build AI that runs marketing — not a product demo, but an honest account of the engineering challenges and the decisions we're still working through.