A longitudinal study of 400+ companies found that despite a 65% rise in AI tool adoption, median PR throughput increased only 7.76% — far below the 3–10x gains often marketed. The core reason: coding represents only ~16% of engineer time, so even significant speed-ups in code generation barely move overall throughput. New bottlenecks like AI-generated code review, skill gaps, and lack of institutional context further limit gains. To improve ROI, engineering leaders should target AI at non-coding bottlenecks (planning, review, documentation), assess codebase readiness before scaling, and measure AI impact across utilization, impact, and cost dimensions.
Table of contents
What the data actually showsWhy the gap existsMore like thisThree steps to unlock more from AI investmentWhere to focus next157 Impressions