Generative AI (GenAI) systems, such as ChatGPT, provide personalized learning opportunities in language education. However, concerns arise regarding their potential negative effects on language competence (Godwin-Jones, 2024; Kostka & Toncelli, 2023). This project investigates the pedagogical potential of GenAI systems in language learning applications. GenAI refers to computer systems that autonomously produce new content (e.g., text, images, sound) using machine learning models trained on diverse datasets (Russell & Norvig, 2020). The study employs a framework by Feuerriegel et al. (2024), categorizing GenAI into three levels: models (algorithm and architecture), systems (user interface with embedded model), and applications (real-world uses). The methodology involved four preparatory steps: identifying AI language learning applications, selecting AI systems, creating prompts, and developing sample texts. Eleven applications were identified, including text generation and revision. For each application, ChatGPT and a tailored system were chosen. Prompts were crafted for AI inputs, often paired with sample texts. For instance, text revision utilized ChatGPT and Grammarly with prompts like “Revise the following text,” applied to intermediate proficiency samples. After inputting prompts into AI systems, the assessment focused on output quality, interface suitability, and the degree of learner autonomy. Results revealed performance variations across applications and systems. Notably, superior AI performance sometimes undermined pedagogical value by limiting learner autonomy. This finding underscores the necessity of balancing AI capabilities with learner agency to enhance language learning. The study highlights the importance of considering how different systems promote or hinder learner agency in developing effective AI-assisted language learning solutions.