Benchmark · 10 questions · about 3 minutes
Two numbers move in opposite directions on most teams. This compares a team against 1,149 professional developers on both, one question at a time.
Why these two. In the same survey, 89% of developers reported a positive impact from AI-assisted code on productivity. Only 25% reported one on the frequency of outages. Speed arrives first. Everything downstream of it arrives late or not at all.
Result
Exposure · how much AI-assisted code the team carries
Control · what happens to it before and after it ships
Across the whole sample, 96% of developers do not fully trust that AI-generated code is functionally correct, and 48% always check it before committing. Distrust is near universal. Checking is not.
The reason is cost, not discipline. 61% agree AI often produces code that looks correct but is not reliable, and 38% say reviewing it takes more effort than reviewing a colleague's work. A pass that returns forty comments is more expensive to read than the risk of shipping. Closing the gap means making checking cheaper, not making it louder.
Toil is the evidence. Developers spend 24% of the work week on it, and that share is identical for heavy and light AI users. It did not shrink. It changed shape: the heaviest users name managing technical debt (44%) and rewriting code AI produced (25%) as their largest sources.
43% of developers are concerned that AI coding tools lack sufficient context about their specific project or codebase. 53% are concerned about a decline in the team's own understanding of the codebase, the second-highest concern in the survey. Both are the same problem measured from different ends.
Hyrax profiles a repository, audits it across security, correctness, maintainability, performance, architecture and operations, submits fixes as pull requests verified against the repository's own tests, and serves that profile live to Claude Code, Cursor and Copilot so they read the current architecture instead of inferring it.