Five Structural Problems of Large-Scale Software in Industry

Friday, April 3, 2026
Schedule : 14:00
Emmanuel Zarpas (MoVe, LIP6)

Large software systems developed in corporations face recurring difficulties that are not isolated defects but structural properties. This talk discusses five such constraints observed in industrial environments: coordination across large teams, process overhead and certification requirements, delivery pressure driven by business timelines, the limits of testing as a proxy for quality, and instability introduced by rapid technological shifts such as AI-assisted development. The discussion is informed by six months of observation of test and build execution in a large code base maintained by roughly 500 engineers, including quantitative analysis of failure patterns in a test suite of more than 100,000 tests. Several of these constraints motivate research questions: how to measure software quality beyond test pass rates, how to reason about large and partially unstable test infrastructures, and how to provide meaningful guarantees for code partially produced by LLMs. The talk aims to make explicit a set of constraints that are central to industrial practice but rarely formalized or studied systematically.

Campus Pierre et Marie Curie, 24-25/405