The article explores how AI is reshaping user interfaces, moving beyond traditional command-based systems. AI interfaces shift from process-driven to outcome-driven, prioritizing user goals over step-by-step instructions. Chat boxes have become the dominant interface due to their intuitive nature and ease of use. However, the inherent "squishiness" of LLMs, where slight prompt variations yield different results, presents challenges. AI products often bury parameters, hindering power users who need explicit controls. The article highlights a "category error" where AI tools, capable of deep creation ("rooms"), are often presented as simple "windows" through the chat interface. This mismatch leads to users improvising, using workarounds to gain control over the AI's capabilities. The solution lies in Generative UI, where AI dynamically chooses the most fitting presentation container for output. Ultimately, the key is to find the right balance between human and machine agency. This involves balancing outcome and process, requiring control and freedom to define future design.