MAGS

Multi-agent governance system. Scenario intelligence across three time horizons with Red Team adversarial testing.

MAGS applies organisational design theory to multi-agent AI governance. Drawing on Phanish Puranam's research across 36 academic papers, it treats the four universal problems of organisational design (task division, task allocation, reward provision, and information provision) as the architectural constraints for a multi-agent system operating in complex regulatory environments.

Specialised agents operate in full isolation during analysis, each applying a distinct analytical framework to the same scenario. A dedicated adversarial agent challenges the emerging consensus by design, preserving the analytical diversity that surfaces genuine uncertainty. Coordination happens only at synthesis, through structured output schemas that enable integration without forcing premature convergence.

The system is built around a human-centric principle grounded in self-determination theory: the humans who review and act on agent outputs must retain autonomy, competence, and relatedness. Without these, oversight degrades into compliance theater.