A Harvard study found that AI coding agents increased code volume but did not improve overall software output because human code review became a bottleneck
Research publication Provisional 55% confidence first seen
Two articles discuss findings and arguments about AI coding assistants. One coverage summarizes a Harvard analysis of large-scale work-event data showing that while AI tools produced more functional code, downstream software output gains were limited as human code review became a bottleneck, leading to longer and more iterative review cycles. The other article provides a critical viewpoint on AI coding’s effects on labor and creativity, arguing that benefits may depend on underlying assumptions.