Descriptions:
Nicole Forsgren—researcher, former GitHub VP of Research and Strategy, creator of the DORA metrics framework, and co-author of *Accelerate*—joins Gergely Orosz at The Pragmatic Summit to discuss what’s actually happening inside engineering organizations as AI coding tools proliferate. Now working at Google on developer productivity and agent experience, Forsgren opens with a paradox that her new book *Frictionless* addresses: AI tools are helping teams produce code faster than ever, yet overall delivery velocity has not kept pace. Her explanation is structural. Bottlenecks that were tolerable at human-scale output—code review queues, deployment pipeline handoffs, security and launch reviews managed by small teams—are now overwhelmed by the volume AI-augmented engineers can generate.
Forsgren describes specific failure patterns she observes across companies. Review automation that previously handled straightforward changes is being removed because AI involvement introduces uncertainty about code reliability, shifting more burden onto human reviewers at exactly the wrong moment. Deployment and release processes designed for a handful of humans making group decisions about candidate builds don’t scale when the number of candidate builds multiplies. She also addresses the measurement problem: the DORA metrics that served as a productivity signal in pre-AI environments are breaking down, and replacing gut-feel with any useful signal is now harder precisely when the stakes are higher.
The conversation also touches on the social dimension of working with AI agents versus human collaborators—Forsgren argues that the honest friction of human collaboration, the kind that caught 100 unnecessary pages in her *Frictionless* manuscript, is something current AI tools cannot replicate and that engineering teams need to consciously preserve.
📺 Source: The Pragmatic Engineer · Published March 22, 2026
🏷️ Format: Interview
