The appetite for generating software architecture diagrams with AI is undeniable, with use cases ranging from solution creation through to documenting existing codebases. Social media is awash with stories where an AI agent has generated diagrams, with most authors claiming that the AI has done "a good job". This generally isn't true though - almost every software architecture diagram I've seen generated by AI is terrible! In this session I'll explain the common problems I see with AI-generated software architecture diagrams, how to fix them, and how to get better results from your AI agent if you do want to use it as a tool to assist with diagram generation.