Abstract: Individual software tools are often well analyzed both academically and commercially. But, developers interact with many, many tools over the course of a day. We constantly build tools for ourselves to make our own development faster, and large development organizations have shared tools that number in the thousands. Large-scale analysis of entire workflows, especially in context of a developer’s day which is filled with interruptions, distractions, and business-critical non-coding tasks is an exciting area. If we understand this area well, we can do prediction and modeling of behaviors outside of individual tools, and we can tackle incredibly interesting problems. These opportunities include reduction of defects through workflow analysis, automatic documentation for even infrequent tasks, UX improvements that span multiple tools, and even predicting outages that impact developers. Machine learning has opened up analyses of data at a scale in this space that were previously too opaque or expensive to consider.

Liane Praza is a Software Engineer at Facebook where she’s worked since 2017 on Developer Infrastructure. She created and currently leads a team evaluating developer workflows, using machine learning to recommend better tools, predict future actions, and find potential problems earlier. She has worked on commercial enterprise software for over 20 years. Prior to joining Facebook she was one of the lead engineers on the Solaris operating system at Oracle/Sun Microsystems. She holds a B.S. in Computer Science with a minor in Philosophy from Purdue university.