Language apps inherit a teaching method designed for Latin, not for spoken fluency. In 1788, Prussia introduced the Abitur, requiring a standardized, gradable language education for universities. Educators adopted the existing Latin teaching method, focused on grammar rules and translation, because Latin was a dead language where oral skills were irrelevant. This "Prussian Method" prioritized reading comprehension and grammatical analysis over conversational ability. When applied to living languages like French and German, this method, later known as Grammar-Translation, persisted. It assumed literacy and fluency were the same, despite evidence like children speaking fluently before reading. Reformers in the 19th century, like Berlitz, advocated for immersive learning, but direct immersion was hard to grade and largely ignored. Linguist Stephen Krashen formalized this distinction in 1982, stating language acquisition (unconscious immersion) produces fluency, while language learning (conscious rule study) only produces test-passing ability. This explains why classroom study often fails to produce conversational fluency. Modern language apps replicate this outdated model, gamifying grammar drills and translation exercises. They optimize for measurable proxies like streaks and vocabulary scores, rather than actual conversational ability. This design choice stems from the difficulty of directly measuring true fluency within an app. However, technology now removes the original constraint of grading at scale with limited resources. AI conversation partners and immersive tools, like Praktika and Langua, are making genuine language acquisition accessible. These products don't invent new pedagogies but leverage technology to enable long-recognized effective methods. The persistence of the Grammar-Translation method highlights how a design decision, made under specific constraints, can become an invisible assumption. It's crucial to reassess if original constraints still exist before optimizing for easily measurable, but ultimately ineffectual, proxies for true user needs.