The Critical Role of Troubleshooting in an ML-based development process

This talk explores the critical role of troubleshooting in modern, ML-driven software development. We first look into traditional DORA metrics and their valuable insights. Nevertheless, their inherent lag and delayed feedback loops present challenges for effective optimization. The presentation introduces “Local DORA” metrics—such as Time To Restore (TTR) a local or a Pull Request failing build— as more actionable proxies that provide immediate feedback, enabling organizations to react swiftly to issues. In particular, optimizing local TTR is paramount for accelerating development speed. The talk will then address the dual impact of AI/ML on troubleshooting: while ML-generated code can complicate debugging issues, AI tools, such as those in Develocity, can significantly enhance troubleshooting capabilities, shorten feedback loops, and therefore improve development efficiency.