Stanford researcher shares insights from large-scale studies on developer output, why early AI productivity claims were overstated, and what engineering leaders should (and shouldn’t) measure when rolling out AI across the software development lifecycle.