How to do Voice of Customer in the Age of AI: A Case Study From My Own Build

I'm putting sunlight on my own process working with AI to figure out voice of customer for PromptKit. Along the way I show you how I think about MVP design and how I use AI to analyze data!

So I launched PromptKit a couple days ago, and I thought it would be a good time to sunlight how we build in the age of AI. The concept of PromptKit is pretty simple—prompts aren’t being treated as seriously as code, so maybe we should build a tool that takes that seriously.

And yes, there are a number of third party tools out there that I’ve looked at that play in the prompt space to some degree, and I think there’s a distinct value wedge here (and room for multiple winners, since org prompting workflows are highly individual).

This note is focused specifically on voice of customer for PromptKit. It’s one of those jobs that would have taken hours or days before and is now possible in just a few minutes with good analysis and an anonymized data set for an LLM.

One of the ideas that’s been rattling around in my head is that there’s a lot of value in just exposing how intensive AI users use AI for real tasks, so that’s some of the concept here. I want you to get an idea for how you could use AI differently by reading this post.

As an aside, it’s somewhat amusing that my entire job earlier in my career is now just done by python tooling in o3 with proper prompting. Wild. Anyway, read on for how we’re thinking of building and a glimpse into how I used o3 to start to get a grip on MVP narrowing for a real AI product.

Oh and sweet tidbit—this is not just vaporware. It’s in code, and so far 60% of code is written by LLM (we’ll see how that holds up as we near launch). Anyway, to get all the juicy deets read on…