Pitfalls in Measuring AI Productivity

How to measure AI developer productivity in 2025 | Nicole Forsgren

Topic 4 of 18

Published Oct 19, 2025 · Clip 12:03–16:00 (3:58)

Nicole critiques common productivity metrics like lines of code as 'a lie,' especially with AI's verbose, gameable output introducing debt. She warns against blindly using old metrics like DORA's speed/stability without adaptation, as AI demands faster feedback loops throughout pipelines, not just end-to-end. SPACE framework adapts better by covering satisfaction, performance, activity, communication, and efficiency/flow, plus new dimensions like trust in AI outputs. Attribution challenges arise: differentiate human vs. AI code for quality, survivability, and bias in retraining loops.