From PR Throughput to Product Velocity: How Dropbox is Rethinking Productivity in the Agentic Era

Last year, many engineering organizations—including Dropbox—focused on adopting AI coding assistants. The results were promising: measurable productivity gains, growing adoption, and a cultural shift toward AI-powered development. But that phase is already behind us.

Today, we're entering a fundamentally different era—one where AI is no longer just a tool, but an active participant across the entire software development lifecycle. In this talk, we'll share how Dropbox is moving beyond copilots to build an agentic SDLC—where AI systems generate code, orchestrate workflows, and increasingly drive development end-to-end. We'll discuss why off-the-shelf solutions weren't enough, and how we're investing in a custom orchestration platform to manage agentic workflows at scale—enabling deeper integration, stronger control, and higher trust.

But with this shift comes a new set of challenges. As AI dramatically increases code generation and pull request volume, traditional systems begin to break. We're already seeing the potential for a scalability crisis—from code review bottlenecks to CI overload—and we'll share how we're proactively evolving our developer platform to stay ahead.

Finally, we'll explore how measuring productivity must evolve alongside the technology. Instead of focusing on surface-level metrics like PR throughput, we're building a deeper understanding of:

• how much code is truly generated by AI,
• how agentic workflows contribute to outcomes, and
• how to measure what actually matters: product velocity—from idea to shipped experience.

This is not a story about tools. It's a story about transformation—how we're rethinking systems, metrics, and engineering foundations to unlock the full potential of agentic development.