AI-Generated Code Waits Nearly Five Times Longer for Human Review as Developer Trust Gap Widens
Analysis of 8.1 million pull requests reveals the bottleneck in software has shifted from writing code to reading it
The numbers paint a stark picture of the emerging friction between AI productivity gains and human verification capacity. Teams with high AI adoption complete 21 per cent more tasks and merge 98 per cent more pull requests, but review time increases by 91 per cent. The acceptance rate for AI-generated code sits at just 32.7 per cent compared to 84.4 per cent for human-written code.
The quality gap is significant: AI-generated pull requests average 10.83 issues per review compared to 6.45 for human code, with logic errors up 75 per cent and security vulnerabilities 1.5 to 2 times more frequent. Change failure rates have risen 30 per cent and incidents per pull request are up 23.5 per cent year over year.
These findings come as AI-assisted coding reaches mainstream adoption. GitHub's Octoverse report found that 41 per cent of all new code is now AI-assisted, with monthly code pushes crossing 82 million. Anthropic's 2026 Agentic Coding Trends Report reported 90 per cent enterprise adoption of AI coding tools.
Developer trust is moving in the opposite direction. Stack Overflow's 2025 survey found that 46 per cent of developers actively distrust AI code accuracy, up from 31 per cent the year before. Only 3 per cent report high trust.
Analysis
Why This Matters
The software industry has automated production but not verification. This asymmetry creates a growing backlog that could slow the very productivity gains AI coding tools promise. It also raises questions about code quality in production systems.
Background
AI coding assistants like GitHub Copilot, Cursor, and Anthropic's Claude have rapidly become standard tools in professional development. The assumption was that faster code generation would accelerate delivery. The review bottleneck was not widely anticipated.
Key Perspectives
Some argue the review gap will close as AI-generated code improves and reviewers develop new workflows. Others warn that the trust deficit is structural — developers fundamentally approach AI code with more scepticism because they cannot infer intent the way they can with a colleague's work.
What to Watch
Whether AI code review tools can meaningfully reduce the human review burden, and whether the 32.7 per cent acceptance rate improves as models get better at matching team coding standards and conventions.