LLMs are très bien at localization

AuthKit's localization process, which translated the tool into 90 languages in five weeks, exemplifies how modern developer tooling and large language models can streamline the translation and localization of digital products. Localization, which adapts digital experiences to align with users' cultural and linguistic preferences, is important beyond language translation as it includes adjustments for dates, numbers, and idioms. The approach taken by AuthKit involved extracting hard-coded strings from their React-based app and leveraging FormatJS for efficient translations. They used tools like negotiator for parsing language headers and implemented a system to generate and manage translation IDs automatically. A major aspect of their strategy was embracing a comprehensive list of user-preferred languages, guided by real user data from headers like accept-language. Developers incorporated a script that uses AI models, specifically Claude Sonnet 4, to automate translations, ensuring structured outputs and deterministic results. The process also included measures for runtime translations and testing across different languages using browser tools. Overall, AuthKit demonstrates how thoughtful design and cutting-edge technology can make localization feasible and scalable, even for complex applications.