Why Software Factories Fail (or: harness engineering is not enough)

# Summary Software factories relying heavily on AI coding agents are failing because engineering loops alone are insufficient for maintaining code quality and reliability. Recent data shows that since widespread adoption of AI coding tools in early 2024, pull request review quality has declined, incidents and bugs per developer have increased, and some companies have experienced outages caused by coding-agent mishaps—contradicting the prevailing narrative that "code is free" and automation can eliminate human oversight.