Listen and watch now on ** YouTube, Apple, and Spotify**.
In this session from DX Annual, Uma Namasivayam, Senior Director of Engineering Productivity at Dropbox, shares how the company’s developer productivity efforts evolved from improving developer experience to preparing for the agentic era.
He explains how Dropbox approached AI adoption across its engineering organization, the impact it had on developer productivity, and why faster code generation is creating new bottlenecks in areas such as code review, validation, and CI/CD. He also discusses Dropbox’s efforts to rethink engineering systems, measurement, and workflows, including the development of agentic tooling and new metrics designed to move beyond PR throughput and toward product velocity.
**Some takeaways: **
Dropbox’s productivity journey started before AI
DXI helped Dropbox identify productivity problems as system problems rather than talent problems. When the company began measuring developer experience in 2023, it found significant variation across teams in DXI scores, PR throughput, and cycle time.Measuring developer experience created a framework for prioritizing investments. The team used DXI to identify friction across areas such as debugging, documentation, and build systems while giving leadership a common language for discussing productivity.
AI adoption required more than access to tools
Dropbox combined executive support, developer segmentation, enablement, and strong guardrails to drive adoption. Different teams and developer roles were matched with different tools and workflows based on their needs.The approach helped Dropbox increase AI adoption from roughly 30% to 100% within six months. During the same period, PR throughput doubled and developer satisfaction with AI tools increased significantly.
Engineers used their extra capacity to tackle neglected work
**As AI increased throughput, engineers naturally pulled maintenance work, migrations, and technical debt from the backlog.Dropbox saw significant growth in these categories without any specific direction from leadership. The additional capacity was often reinvested into engineering health.**Teams used the opportunity to address long-standing issues that had accumulated over time rather than focusing exclusively on new feature development.
The next challenges are scale, trust, and measurement
**Dropbox believes the move to agentic engineering creates three major challenges: scale, validation and trust, and measurement.Existing development systems were not designed for a world where AI dramatically increases code throughput. As code generation accelerates, bottlenecks are shifting toward code review, validation, and CI/CD systems.**The company is already seeing pressure move downstream in the software development lifecycle.
Agentic engineering requires redesigning the entire system
Uma compared the transition to the shift from steam-powered factories to electric factories. The biggest gains came from redesigning the entire system rather than simply replacing one technology with another.Dropbox is investing in agentic workflows across the SDLC and building Nova as an orchestration layer. The company is evaluating roughly 30 development steps, and one in twelve pull requests is already being generated by Nova.
PR throughput is becoming a less useful measure of productivity
Dropbox believes traditional engineering metrics need to evolve alongside AI. As agentic workflows become more common, measuring productivity through pull request volume alone provides an incomplete picture of engineering output.**The company is increasingly focused on metrics such as AI contribution, loaded cost per PR, agentic workflow coverage, work distribution, and time to ship.**The goal is to better connect engineering activity to customer value and business outcomes.
In this episode, we cover:
([00:00](https://www.youtube.com/watch?v=w0kHCjTOvyo)) Intro
([00:57](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=57s)) The beginning of Dropbox’s DX journey
([02:34](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=154s)) AI adoption at Dropbox: what made it work
([04:46](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=286s)) The results of Dropbox’s AI adoption efforts
([05:39](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=339s)) What the results mean for the business
([06:55](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=415s)) The phases of AI adoption and where they are now
([08:00](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=480s)) The new bottlenecks
([09:16](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=556s)) Three challenges Dropbox faces moving into agentic engineering
([10:05](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=605s)) How Dropbox is redesigning the SDLC for agentic engineering
([15:46](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=946s)) The new metrics that matter
([19:16](https://www.youtube.com/watch?v=w0kHCjTOvyo&t=1156s)) Final takeaways
Referenced:
• [Dropbox](https://www.dropbox.com/)
• [Developer Experience Index (DXI) | DX](https://getdx.com/developer-experience-index/)
• [DX Core 4 Productivity Framework](https://getdx.com/corefour)
• [Cursor](https://cursor.com/)
• [Claude Code | Anthropic’s agentic coding system](https://www.anthropic.com/product/claude-code)
• [Jira | Project Management for the AI Era | Atlassian](https://www.atlassian.com/software/jira)
• [GitHub](https://github.com/)