This document summarizes a presentation given at the Global Data Science Conference 2017. The presentation discusses challenges with production machine learning/AI systems, including models degrading over time, lack of scalability, and lack of integration. It provides top ideas to address these problems, such as using a DevOps cycle, decoupling APIs and models, using schedulers and containers, automating processes, and potentially rewriting parts of systems. The presentation emphasizes the need to focus on keeping models running effectively in business/production settings.