We are collecting incredible amounts of data from professional software developers doing their jobs. Can we leverage this data to make developing software easier? Recent advances in deep learning—particularly around “relational deep learning” methods that can process irregular, graph-structured data—are giving us a powerful new toolbox to tackle the hard problems in this space. Danny will describe how we’re leveraging these tools to build systems that learn to edit professional developer code in order to resolve small programming errors. He’ll also discuss broader outlooks for where this research area is going and point out some really hard problems in this space for which we don’t currently have good answers.