AI and Software Quality Guardrails

Your Code as a Crime Scene: The Psychology Behind Software Quality - Adam Tornhill

Topic 8 of 10

Published Dec 1, 2025 · Clip 42:07–54:31 (12:24)

Adam argues AI's impact is up to us—it can help but risks becoming a technical debt generator at machine speed. Writing code is only ~5% of a developer's week while understanding code is ~70%, so AI optimizing typing at the expense of comprehension is dangerous. Guardrails include code health checks in the IDE (discard unhealthy AI code), security scanning, questioning whether code is needed at all, and reemphasizing reading skills and practices like TDD, code reviews, and pair programming. He shares a cautionary tale of a company that scaled an AI tool to thousands of developers and saw production bugs, discusses how healthy human-readable code is also AI-friendly code (per Thoughtworks' Tech Radar), and endorses documentation capturing the 'whys' and trade-offs as valuable for both humans and AI.