Note
May 14, 2026
2 min read
Codex, Product Taste, and the Problem of Too Much UI
By Cristiano Pierry
A short reflection on where AI coding tools still need stronger product taste, visual hierarchy, and restraint.
OpenAI Codex has become one of my most useful tools for building quickly.
But one area where I still see a gap is design sensitivity.
Small example: I was creating an AMA section for my personal website. The page was meant to introduce “Bob,” a public AMA assistant that answers questions about my writing, profile, and public notes.
The first version Codex produced technically worked, but the design repeated “Bob” everywhere. Bob in the badge. Bob in the headline. Bob in the card. Bob in the status. Bob in the body copy.
It was functional, but not refined.
To be fair, some of this is on me. A better prompt probably would have specified the hierarchy, repetition tolerance, and desired level of visual restraint more clearly.
But I do notice this more often with Codex than with some other LLMs: it is very strong at implementation, architecture, and getting the thing to work, but less naturally sensitive to product taste, visual hierarchy, copy restraint, and the little design judgments that make something feel polished.
This is a reminder that “working” and “designed” are not the same thing.
The next unlock for AI coding tools is not just better code generation. It is better product judgment.
Because the real goal is not to generate more UI.
It is to generate less unnecessary UI.
This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.