Klaviyo Composer
Klaviyo / Product design
I'm a founding member of Klaviyo’s Composer team. I work on AI interactions through user research, design, and now code.
My work on Composer
I joined Composer as a founding team member. My work covers interaction design and user research. These days, I also work directly in the code.
Composer brings AI into marketing work, including creating campaigns and flows from a prompt. I focus on how people direct the AI and edit what it creates.
Introduce AI where it can help
We worked on a balance between letting people discover Composer organically and showing where it could help within their existing workflow. Repeated tasks were a starting point: where could AI save someone time on work they already do every day?
Give the workflow room
I pushed back on proposals for large, prominent input bars. They would take over the screen, especially for business owners and enterprise customers working on small Windows laptops.
I wanted the introduction to match the premium experience we promised. Getting people to try a feature was only useful if that first encounter gave them a reason to keep using it.
Help people know where to start
We couldn’t assume people already knew what to ask an AI. The technology could feel intimidating, even when it could help with a familiar task. We wanted to make those opportunities clear while keeping people in control of what happened next.
Keep track of what AI is changing
I explored object widgets that would keep the items being changed visible. Someone could move into a preview, return to their previous work, or switch between objects without losing track of what they were editing.
The editing state needed to be explicit. AI could add a layer to the workflow, but people still needed to understand where they were and how to get back.
Make past conversations easier to find
In an October 2026 research session, a participant couldn’t find a previous conversation and looked for “Past chats.” After hearing similar feedback on other calls, I recommended returning chat history to the sidebar and agreed to take on the change.
Design through code
I try interactions directly in the implementation, using AI coding tools to make changes and refine how they work.
From shared edits to AI changes
In Universal Content, one edit could affect many messages. I’m working through a related question with AI: how do we show the reach of a change before someone commits to it?