Why AI Struggles at Front End Development

Article Summary

Adam Argyle at nerdy.dev examines why current AI tools fall short for frontend development, citing four structural causes: training on outdated, template-heavy code; inability to render or perceive UI output; lack of understanding of why CSS and layout decisions are made; and no awareness of the dynamic browser environment. The piece acknowledges AI handles scaffolding and tedious migrations well, but consistently underdelivers on bespoke layouts, custom interactions, accessibility, and performance — with quality degrading as component complexity grows.

Discuss on HN

Discussion Summary

• Sharp divide in the comments: backend/ML engineers report AI works great for frontend (“it has really enabled me to build visual apps”), while experienced frontend devs find it lacking for anything beyond scaffolding
• Practical workaround shared: use ImageMagick to compare screenshots against mockups, targeting <5% difference — one commenter claims to have built an entire website in 2 days with this technique
• “If AI really sucked at front end I’d have a job right now” — a poignant counterpoint suggesting AI is already good enough to displace some frontend roles, even if imperfect
• Several recommend creating CODING.md/FRONTEND.md files with rules and expectations to get AI 80% of the way there, then focusing human effort on the design-critical 20%
• Deeper insight: AI struggles with frontend partly because frontend itself churns — frameworks, paradigms, and best practices change so fast that even humans don’t always know “why we do things”