Keep agent-built screens ready for 8 languages

You lead a small team rebuilding a legacy web app's UI in React, and engineers generate components with AI agents. The new UI must ship in 8 languages from day one, including Arabic, which reads right to left.

Components arrive fast. But review keeps finding the same problems: hardcoded English text, sentences joined from pieces, and dates and prices formatted by hand.

How do you organize the work so translation holds across every component, and where do people have to check?