How to measure AI developer productivity in 2025

Nicole Forsgren is a researcher and partner at Microsoft Research who leads developer productivity research, co-authored the State of DevOps reports and the book Accelerate, and helped create the DORA and SPACE frameworks. The conversation covers how to define, measure, and improve engineering productivity and developer experience, including DORA metrics, the SPACE framework, benchmarks, common pitfalls, and the impact of AI on engineering work.

01Key learnings

• Before measuring anything, get crisp on the problem or goal, since teams often interpret vague mandates like 'improve developer experience' in very different ways.
• Shipping smaller changes more often tends to improve both speed and stability together, because small batches have smaller blast radius and are easier to debug and restore.
• DORA's four metrics (deployment frequency, lead time, time to restore, change fail rate) give a benchmark-style view of where you stand, but the real value is knowing where you are and making progress.
• Treat productivity as a holistic measure that includes satisfaction and wellbeing, since Nicole finds these are highly correlated with other productivity outcomes and early warnings when they decline.
• When choosing metrics using SPACE, pick at least three metrics across different dimensions and keep them balanced, so optimizing one measure doesn't break another.
• Combine data from systems with data from people, since surveys can reveal hidden toil and gaps (like untracked code) that instrumentation alone will miss.
• Use the four-box framework to link words to data: state the outcome in plain language, define measurable proxies, then validate with stakeholders before running correlations.
• Make your work accessible to your key audience by translating findings into a few sentences tailored to their role and vocabulary.

02Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.