The Velocity-Quality Divergence in AI-Augmented Software Engineering: A Staggered Difference-in-Differences Evaluation of Governed Downstream Repositories

Working draft of Chapters 1 and 2 of a Doctor of Business Administration (DBA) dissertation in progress at ESGCI Paris. The thesis evaluates whether AI-assisted software engineering delivers a durable net productivity gain once downstream review, revert, and defect-repair costs are counted, or whether observed throughput gains are offset by a quality wedge borne by governed downstream repositories. Chapter 1 sets out the research problem, the objectives, and the four hypotheses on velocity, quality, cost, and review externalities. Chapter 2 is the literature review, organised problem-first around the productivity paradox, the task-based model of labour, absorptive capacity, and the empirical evidence on developer productivity, code quality, and reviewer load in AI-augmented workflows. The empirical strategy is a staggered difference-in-differences evaluation on GitHub Copilot adoption in corporate-sponsored open-source repositories, using the Callaway and Sant'Anna (2021) estimator on public GitHub Archive data. The identification strategy, sampling frame, and estimation are drafted at blueprint depth for the next submission (Chapter 3, Research Design and Methodology) and are not included in this deposit. This is a living document circulated for supervisor review and peer feedback. Content is subject to substantive revision. Please cite the concept DOI so the reference always resolves to the latest version.