The share of AI-generated code changes reaching production without a separate manual review step has increased significantly over the past six months, according to new data from Cursor reported by Business Insider. Cursor, the AI coding tool that has become one of the fastest-growing developer platforms, says AI-generated code is also surviving in production at higher rates than before.
What the Data Shows
Cursor does not directly measure the quality of fully autonomous code, according to Business Insider’s Alistair Barr. What it does track is how much AI-written code makes it to production and how long it survives there. Both metrics are trending upward. The implication: developers are increasingly comfortable treating agent-generated changes as production-ready without routing them through a human reviewer first.
This builds on a trend Business Insider reported in March, when data showed the AI coding boom was producing more shipped software with little measurable hit to quality. Six months later, the shift has moved from “AI code is good enough” to “AI code doesn’t need a second pair of eyes.”
The Governance Gap
The pattern raises a practical question for engineering organizations: is developer trust outpacing institutional controls? Most software teams still have formal review policies that require at least one human approval before code reaches production. If individual developers are routinely bypassing that step for AI-generated changes because the output seems reliable, the policy exists on paper but not in practice.
This is not a theoretical concern. Code review serves functions beyond quality assurance. It distributes knowledge across a team, catches architectural drift, flags security patterns, and creates an audit trail. When a human reviews code, they are not just checking “does this work” but also “should we be doing it this way.” An AI agent generating correct code that violates team conventions or introduces subtle dependency risks will pass the “does it work” test while failing the broader check.
Production Trust Without Production Guarantees
Cursor’s data point lands in a broader context. Harvard Business School and INSEAD research published last week found AI-native startups operate with 25% fewer employees and comparable valuations, suggesting that smaller, faster teams are already building around agent-generated output. The question is whether the governance scaffolding that made code review a standard practice scales down with the team, or simply disappears.
For organizations shipping AI-generated code at volume, the immediate task is not to slow adoption but to instrument it. If review gates are being skipped, the alternative is automated testing, deployment monitoring, and rollback systems that catch what a human reviewer would have caught. The Cursor data suggests developers have already made the trust decision. Organizations that want to maintain control need to catch up.