
Output vs Outcome with AI

Building Your Software Factory | Benedikt Stemmildt
Topic 3 of 11
Published Oct 29, 2026 · Clip 11:49–17:11 (5:22)
They discuss how individual engineers getting faster with AI doesn't automatically make companies ship faster: more commits and PRs lead to pipeline bottlenecks in reviews, validation, or requirements. Benedikt observes that backlogs can drain and teams start building features that make products worse because they can. He argues the key question is how to spend the time gained from AI efficiency—on features, validation, hypothesis-driven work, or cleanup—noting that only about 20% of his agents build features while 80% does refactoring and architectural improvement.