This document discusses building usable AI tools and challenges related to machine learning applications. It notes that uncertainty, explainability, and lack of training data need to be addressed as user experience problems. Ensuring predictions come with explanations to build trust is important. Techniques like LIME are being developed to explain neural networks. Getting humans involved in providing data through techniques like transfer learning and human-in-the-loop can help address the training data problem. The document distinguishes between tools for doing AI work and intelligent conversational agents.