A registered team-randomized test of individual versus manager-and-team AI adoption incentives

Source checked September 30, 2026: Johannes Berndt, Raffaella Sadun, Soumitra Shukla, Florian Englmaier, Nikolaus von Hesler, Jorge Tamayo and Simon Kayser, AEA RCT Registry AEARCTR-0017768, Incentive Structures for AI Technology Adoption: A Cluster-Randomized Trial in Engineering Organizations, first published February 4, 2026. This is a preregistration marked “In development,” not a results paper. Siemens and Harvard/LMU researchers plan a three-arm team-cluster randomized rollout of an AI engineering assistant used for code generation, interface design and hardware configuration: control (~500 teams), individual incentives (~500), and manager-plus-team incentives (~500). Planned ~1,500 teams, ~2,000–4,500 engineers (wide cluster-size range). Intervention dates listed February 1–April 30, 2026; listed study end December 31, 2027. No treatment effects are reported in the registry as checked.

What it can test, if completed and published. Relative differences in incentives around experimentation and knowledge diffusion across teams, with weekly active use, feature breadth and sustained use (at least weekly use in weeks 9–12) as primary outcomes. Secondary measures include perceived productivity, telemetry on task proxy time (compile intervals/errors and project opens/closes), discussion/teaching frequency, and intended future use; the description mentions collecting data three months after treatment. The contrasts confound several ingredients: manager involvement, team incentives, and individual-versus-collective framing are not separately randomized. A positive effect on telemetry would not establish customer benefit, product quality, maintained-code economics, or that a useful shared team artifact caused the effect. No agent-specific accepted-changes, review-time, defects or repair metrics are preregistered as outcomes.

Why it matters here. Directly addresses Dru’s September 29 hypothesis at the proper randomization unit (team) but tests incentives rather than his precise manager-attended onboarding-plus-shareable-outputs strategy. Put on a future results watch; do not claim that it confirms manager sponsorship. Related: team adoption question; usage is not value; missing value and cost terms.