Most engineering organizations now have access to the same frontier models and coding agents, yet their outcomes diverge dramatically. Why do some teams ship low-cost, low-rework AI-generated code with minimal human intervention, while others grind through clarification loops, review churn, and runaway token spend?
Drawing on new research analyzing real-world agent trajectories across hundreds of engineering teams, plus market survey data on AI adoption, spend, and proficiency, this talk examines what distinguishes high-performing teams. The evidence points away from model selection and toward measurable, trainable factors.
Attendees will leave with a practical framework for instrumenting their own organization's AI effectiveness, including leading indicators they can measure before costs land, concrete practices that improve outcomes, and a clear-eyed view of where the market stands.
