Empirical AI Productivity Gains and Tools

How to measure AI developer productivity in 2025 | Nicole Forsgren

Topic 12 of 18

Published Oct 19, 2025 · Clip 32:17–36:36 (4:19)

Gains are real but hard to measure; focus on end-to-end velocity (idea to experiment/production). AI enables rapid prototyping; lines-of-code rises (double for heavy users via unblocking), but quality/survivability key. Tools excel at bug-finding (e.g., Codex on gnarly issues), unit tests, docs—agents need good data/comments for best outputs. Documentation cleanup boosts AI performance; tools like Devin aid this loop.