We do already implement this for Assistant, and any other AI features in the core Search product; those all use the same backend. This ticket is specifically about Translate, which is its own codebase/infra apart from all of our mainline things - but a great reminder, thanks.