Thinking
The judgment behind AI builds
This site was built in conversation with AI, plus code I refined. These are the decisions I made along the way, with the commits that prove them.
- Stack
Writing
Notes on designing with AI
What I'm noticing as AI takes the execution: where judgment moves, and what designers still own.
AI didn't make design faster for me. It made it denser with decisions.
This week I made 96 design iterations on one project from 0 to 1. The production layer is mostly gone. What's left is the part that was always the job: deciding.
Notion AI deleted my data and asked me to pay to restore it.
I asked it to help delete one column. It deleted several I never mentioned. My first time being this frustrated while building with AI, and a lesson in what an agent owes the user before it acts.
As code becomes cheaper to write, deciding what to write becomes the valuable part.
A line from Catherine Wu, Head of Product for Claude Code, that sat with me for a week. As AI speeds up execution, the real differentiator becomes product judgment.
After 5 years and ~$1,300, I'm finally moving off Webflow.
It was the right call for a long time. With Claude Code, Cursor, and Vercel, I can build and ship from the terminal, with less overhead and more room for the decisions that matter.
AI creates perfect spacing. It doesn't see the optical illusion that makes it feel wrong.
Mathematically consistent, system-wide, and still off. Noticing that takes human attentiveness, which is why designers stay crucial: not just research or knowledge, but attention.
Designers aren't supposed to set up local dev environments.
But the design-to-engineering handoff has been broken for years. We can see something is off and not why. Running the code locally closes that gap.
Proud to have led design on this launch from 0 to 1.
Remitly's conversational experience, a step toward meeting customers where they already are. The design thinking behind it is in the case study.
Go deeper
See how this thinking holds up at product scale.