At Riverside, I was leading a team making core experience changes to how creators organize their content. We’d identified two kinds of creators through qualitative research, but before making changes to tens of thousands of customers, we needed to quantify what people were actually doing on the system. The most important thing to understand was how many studios each creator had and how many recordings per studio. I copied this information into my copilot, which already had a lot of context about Riverside. I attached data showing each recording, account, studio, and how they all related. What’s really cool is Claude not only processed the data but also wrote code to visualize it. It decided to do a scatterplot - this was AI’s decision, not what we did at Riverside. You can see it’s front-end code, which is pretty valuable in itself. But I had some feedback based on hindsight. The analysis at Riverside was done by Nir Einav, one of the sharpest product analysts I’ve worked with. He had the idea of doing a heat map. So I asked Claude to do a similar visualization with the same data, but flip the X and Y axes and use a heat map instead. When we did this in real life, we got a really clear sense of those same groups we saw qualitatively in interviews. It quantified exactly how many people were in those groups because we could see the clusters. We didn’t have to put in any rigid criteria - it was very visceral. I could give more feedback, like asking for a heat map that uses red, yellow, and green colors. What’s cool about using HTML and CSS to visualize is that it can quickly iterate. The data and patterns remain the same - it’s just changing the front-end visualization. It’s really easy for AI to write code and iterate on visuals. I could also ask it to make the numbers more readable and keep refining. This could be a prototype that I can then take to a product analyst to finish. ➡️ LLMs excel at rapid visualization prototyping. They can generate and iterate on HTML/CSS visualizations in seconds, letting you explore different ways to see patterns before involving your data team.