Researchers analyzing granular engineering data from over 700 firms found that AI coding agents increased total lines of code by 30 percent and commits by 20 percent, but did not statistically improve the resolution rate for software features. The efficiency gains in the coding phase were absorbed by a 49 percent increase in human code review time, as reviewers requested changes nearly twice as often and left more comments. Consequently, the study detected no significant change in employment levels or net software output attributable to AI adoption. For financial institutions, this evidence suggests that near-term enterprise ROI from AI coding tools may be lower than projected, as productivity gains are offset by downstream review bottlenecks rather than translating into headcount reductions or accelerated delivery.
56 other tracked developments touching the same claims.