Why Your AI Agent Framework Is Leaking LLM Logic Into Your Code

Current AI agent frameworks lack proper context abstractions, forcing developers to embed LLM-specific logic directly into business code, creating unnecessary complexity and maintenance headaches. Mozaik addresses this by introducing rich domain models for context management based on the OpenResponses specification, enabling clean separation between application logic and LLM concerns. This shift could fundamentally reshape how developers architect scalable AI applications while keeping codebases maintainable and focused on their core domain.