
Large Change Sets and Non-Determinism

AI Coding Workflows 🛠️ — with Birgitta Böckeler
Topic 4 of 13
Published Sep 29, 2025 · Clip 10:52–15:31 (4:39)
Birgitta explains that large change sets increase deployment risk regardless of how they're created, connecting to continuous delivery principles like small, frequent deploys. She argues non-determinism is inherent to LLMs—"hallucination is the key feature"—so there will always be non-zero probability of unwanted output, including from human specification gaps. Larger workflows raise the probability of hard-to-detect errors, so testing strategies and reviewing experiences must evolve if large-scale AI generation is desired.