Observation that individual AI-assisted developers report speed gains that do not cleanly show up in team-level velocity—explained by review load, rework, and context-switching costs.
The AI productivity paradox echoes the 1980s IT productivity paradox: faster code generation shifts bottlenecks to review, integration, and debugging. Measurement needs team-level metrics like defect escape rate and code churn, not just PR throughput.