Analysis Questions AI's True Impact on Delivery Speed
While AI coding assistants can accelerate code generation, they often amplify existing organizational bottlenecks rather than improving end-to-end delivery speed. In a podcast discussion, RiverGlide founder Antony Marcano argued that most organizations' performance is harmed by AI-augmented coding, suggesting a new fifth DORA metric, "rework rate," to measure its true impact on quality.
- The four original DORA metrics, developed by Google's DevOps Research and Assessment team, measure software delivery performance by tracking velocity (Deployment Frequency, Lead Time for Changes) and stability (Change Failure Rate, Time to Restore Service). - Before the "rework rate" proposal, the official DORA team had already added a fifth metric, "reliability," in 2021 to provide a more holistic view of operational performance and service health. - Corroborating the concerns about quality, Google's 2024 DORA report found that while AI adoption led to minor gains in code quality (+3.4%) and review speed (+3.1%), it was also associated with a 7.2% decrease in delivery stability and a 1.5% drop in delivery throughput. - Studies show a significant gap between localized coding speed and end-to-end delivery improvement; one analysis by Thoughtworks estimated that while AI assistants made individual coding tasks about 30% faster, the overall improvement to cycle time was only 8-15% due to downstream pipeline constraints. - This discrepancy arises because AI-generated code increases the volume of pull requests, which then floods existing bottlenecks in manual code reviews, testing environments, and deployment processes, slowing the entire system down. - Antony Marcano, the co-founder of RiverGlide, has a background of nearly three decades in software development, with a focus on agile coaching and software craftsmanship methodologies like Behavior-Driven Development (BDD) and Test-Driven Development (TDD), which prioritize quality and sustainable pace. - Research on AI's impact has yielded conflicting results depending on the context; a 2023 lab study by GitHub and MIT showed a 56% speedup on a specific, isolated coding task, whereas a 2025 METR study of real-world open-source projects found tasks took 19% longer with AI assistance.
