Machine learning is a hot topic almost in every domain right now, including software engineering. More and more research papers are being published, introducing new approaches and improving existing ones, solving dozens of different tasks in this domain. Many of these papers report very high quality of their models, so sometimes reading them leaves you wondering, why don't we see features like this in our development tools? Well, in most cases, it is not that straightforward. In this talk, I discuss the way research ideas and academic prototypes find their way into software products at JetBrains. I will show several cases of how machine learning approaches are integrated into a modern IDE used by millions of developers every day. I will also highlight the challenges we usually face while doing it and discuss the promising research areas in this field.