Hey Reader,

AI has gotten surprisingly good at reasoning through design problems, the part that still falls short is usually the specificity. Without references, it’s mostly working from assumptions. And that’s not how most of us design in practice.

That’s what my newest video is about.

I’m kicking off a new mini-series with Mobbin on why AI-assisted design often falls short and how better references can lead to better outputs.

In this first episode, I walk through an example in Claude. The first version is thoughtful, but it’s still missing something important: evidence from real shipped products.

​

Instead of asking AI to design from scratch, I ask it to first study real shipped screens, then use those references to inform the next iteration. It doesn’t replace design judgment, but it does give you much better inputs to work from.

​Watch here​

I’m going deeper into the workflow in episode two, what would you want to see tested with Mobbin MCP?

Hit reply with a flow or product moment you’d be curious about.