Hi, thanks for offering this Q&A!
We've always used real translators for our apps, but recently our vendor went out of business, and this WWDC is focusing heavily on agentic localization. It's clear this is where the industry is trending, but we're worried about our user experience if we have a fully AI-driven localization.
How do you recommend verifying that agentic localization output is correct, friendly, appropriate, etc?
Hey! Sorry to hear about the vendor, that's some unfortunate disruption for you.
Verification is definitely one of the biggest challenges here, but we have a couple of things to suggest (some of which we also cover in our session, Translate your app using agents in Xcode).
When it comes to finding functional issues for other languages (truncation issues, clipping issues, etc.), agents have the ability to check a lot of their work using the tools baked right into Xcode, like the ability to render a SwiftUI preview of a view in another language, or to launch and interact with your app in the iOS simulator in another language.
As for translation quality, you have a couple of paths you can take. We strongly recommend involving native speakers as testers of your app, either through TestFlight or through any other testing process that you might have. Another thing you can try would be to use different agents from different model providers to perform "adversarial review"—e.g. you have Claude Agent do your initial translation, and then you ask Codex to review the translations based on all of the context that it can pull about the strings and the code that contains them.
Alternatively, if you still want to keep humans in the loop, you could invest in vendors to QA your translations (instead of creating them in the first place).
I hope this provides you a bit of clarity on the direction you'd like to take!
Avery
