One weird trick for less sloppy AI frontends: ask for Qt style

A developer experimenting with AI-generated personal software ran into a familiar problem: no matter what visual style they requested, the output carried the same generic ‘AI slop’ feel. After repeatedly regenerating a simple single-page app in different styles, they concluded that slop isn’t itself a style — it’s a layer of genericness that agents smear over whatever aesthetic you ask for. Requesting a page that looks like X reliably produced X-plus-slop.

The exception was asking the agent (gpt-5.5-thinking via Codex CLI) to make the app look like a Qt desktop application. To the author’s self-described tasteless eye, that single instruction stripped out nearly all of the slop feel, and the trick generalized: after converting their other personal tools to ‘Qt style,’ they found them all noticeably improved. The test app itself was an interactive 270-to-win electoral map built from an Axios forecast of 2030 electoral college shifts, generated by dumping a screenshot into ChatGPT.

The post is less a polished finding than an open question for the vibe-coding era: are there other design vocabularies — likely ones with strict, well-documented conventions like Qt’s — that models can reproduce without layering on generic AI aesthetics? The author explicitly invites others to experiment, suggesting that constraining agents to established design systems may matter more than describing the look you want.