When coding agents translate your app: what breaks, what to keep

AI coding agents now add translation keys and draft the translated strings as part of a code change, with more context than most translators ever get. What they do well, what does not carry over time (consistency across releases, review and provenance, terminology governance, QA over the whole corpus), how teams handle it today, and the agent-in-the-loop pattern that keeps the speed without losing the record.