994: AI Sucks At CSS

In this episode of Syntax, hosts Wes and Scott dive into the growing frustration around AI's poor performance with CSS, despite its strengths in backend code. They explore why AI-generated designs often look bland, overly similar, and lack visual sophistication—attributing this to model biases, over-reliance on generic system prompts, and a lack of true design taste. The hosts emphasize that while AI excels at solving deterministic problems, it struggles with the subjective, creative aspects of front-end development, especially when it comes to layout, typography, and visual hierarchy. They discuss real-world examples, such as AI failing to use modern CSS features or creating bloated, one-off classes, and stress that human expertise and taste remain irreplaceable. The conversation also touches on broader implications: how AI is reshaping developer workflows, the importance of understanding code even when using AI, and the need for better tools to debug performance issues. The episode concludes with a discussion on emerging tools like VibeRules and skills management, cautioning against over-reliance on external packages due to security risks like prompt injection.