From PR throughput to product velocity: How Dropbox is rethinking productivity in the agentic era

The podcast episode introduces the theme of shifting focus from traditional PR throughput metrics to overarching product velocity, emphasizing Dropbox's three-year journey to improve engineering productivity. It highlights the importance of learning from industry convergence and sets the stage for discussing Dropbox's internal transformation.

This section details Dropbox’s strategic investments starting in 2023 to enhance developer experience through infrastructure, workflows, and team upskilling, recognizing system issues over talent gaps. It discusses initial metrics like DXI score, PRs per engineer, and cycle time, and describes how these metrics improved with concerted effort. The narrative then shifts to Dropbox’s AI adoption starting in early 2025, driven by top-down executive support, process adjustments, developer segmentation, enablement, and safety protocols, leading to a full adoption of AI tools and a doubling of PR throughput.

The discussion explores the dramatic increase in AI-assisted code generation, with Dropbox reaching up to one in 12 PRs produced by Nova, their orchestration layer. It explains how AI acceleration introduces new bottlenecks in code review, validation, testing, and CI/CD processes. The shift results in increased review and validation times, higher CI/CD costs, and cognitive overload for developers, necessitating a redesign of workflows, metrics, and system architecture to keep pace with speed and quality.

Dropbox is rebuilding its entire software development lifecycle to be agent-centric, focusing on orchestrating workflows across systems rather than optimizing individual tools. The development of Nova as an orchestration platform exemplifies this approach, connecting various parts of the SDLC via a unified agent system that automates and scales processes like code review, migration, and testing. This section emphasizes the importance of system-wide metrics and new ways to measure value—shifting from PR counts to customer-centric impacts.

Traditional metrics like PR throughput are no longer sufficient in an agentic SDLC. Dropbox is adopting holistic, product-oriented metrics such as AI contribution to PRs, code lines activated by AI, system costs, work distribution, and agent coverage. The focus is on impact, ROI, and customer value, alongside quality measures like defect ratios and rework rates. These metrics aim to provide a clearer picture of how AI accelerates product delivery and enhances value.

The episode concludes with practical advice: anticipate increasing PR volumes, identify bottlenecks beyond code generation, invest in validation and trust systems, and run development as a product with strong measurement frameworks. It stresses the importance of building connective tissue—integrations and orchestration that connect systems for scalable, reliable AI-driven engineering. The overarching message advocates continuous learning, industry collaboration, and deliberate system redesign to thrive in the agentic era.