Software Teams Face 57.9% Incident Jump, Urge Risk-Based AI Code Review

2 articles · Updated · O'Reilly Media · Aug 20

Production incidents tied to AI-heavy coding are climbing fast: monthly incidents are up 57.9% year over year, and PRs merged without any human or agentic review have risen 31.3%.
Faros AI, CodeRabbit and Lightrun data point to the same pattern—AI-coauthored PRs carry 1.7 times more bugs than human-written code, while 43% of AI-generated changes need debugging in production.
The report argues the deeper failure is “cognitive debt”: teams ship faster than they understand, then lose the ability to make safe architectural decisions or respond quickly when incidents hit.
Its proposed fix is risk-based routing: require line-by-line review for security, money movement, permissions and destructive data changes, but allow lighter systems inspection for low-risk, well-tested code.
One rule stays constant across tiers: the agent that writes code should not be its only reviewer, as engineering leaders are urged to phase in governance, tooling and shared comprehension over 6 to 12 months.