This project aims to develop self-optimizing AI agents for educational content generation. Recent advances in Large Language Models (LLMs) have made it possible to automatically generate online courses, including lessons, assessments, and supporting materials. However, the quality of generated courses still depends heavily on manually designed prompts and workflows. The goal of this project is to investigate how educational content generation agents can automatically improve themselves over time. Inspired by evolutionary algorithms, the project will explore methods for optimizing agent prompts, skills, and workflows using quantitative measures of course quality. A key challenge is developing automatic evaluation metrics that can assess pedagogical quality directly from generated content, without requiring student studies. The project will therefore study metrics related to curriculum coherence, learning objective alignment, assessment quality, and instructional design, as well as recent LLM-based evaluation approaches. The expected outcome is a framework for self-improving educational AI systems that can generate higher-quality learning materials with minimal human intervention. By making high-quality course creation more scalable and accessible, this research has the potential to expand access to education across a wide range of subjects and learners. This project is in collaboration with LearnPack (https://learnpack.co/)