Why AI is generating ‘lowest common denominator’ React code

Artificial intelligence coding assistants are churning out React code, but there's a fundamental problem: they're learning from some of the worst examples on the internet.

Seth Webster, executive director of the newly created React Foundation, says large language models have been trained on publicly available code that represents "the lowest common denominator." Meanwhile, the best coding practices remain locked away in private repositories, invisible to AI training systems.

"They're trained on the worst Svelte, they're trained on the worst Swift, because what they're training on is publicly available code," Webster explains. The result? AI that performs like a middling engineer, not the coding virtuoso many hoped for.

Take Claude's habit of using refs to track state in React — it's not terrible, but it's far from ideal. Webster says models don't understand that business logic should live in external services, not crammed into React components.

The React Foundation now aims to improve AI-generated code through better evaluation systems. Until then, Webster warns, AI "requires a lot of guidance, and it will for a while to come."