Agentic AI at Scale: Redefining Management for a Superhuman Workforce (sloanreview.mit.edu)

MIT Sloan Management Review and BCG convened an international expert panel and ran a 1,221‑respondent executive survey to probe accountability around “agentic AI” — systems that pursue goals autonomously, learn and adapt, and act at superhuman speed. A clear majority (69%) say this autonomy, opacity, and scale create a governance gap that demands reimagined management: explicit agent roles, permissible actions, data guardrails, confidence thresholds, and continuous oversight. A strong minority (25%) push back, arguing existing management practices can be adapted and that accountability must remain squarely with people and organizations, not the software. Technically, agentic systems combine memory, reasoning, and adaptive learning, making causation and fault hard to trace and increasing the risk of fast, cascading errors. Panelists recommend life‑cycle–based governance with recurring technical audits, automated monitoring with ethical guardrails, traceability and audit trails, and clear human roles and escalation paths. They also urge defining contexts where AI‑led decisions are acceptable versus those that require human intervention. The net implication for AI/ML teams and managers: embed continuous, technical oversight into operations, engineer systems to support human supervision (e.g., thresholds, explainability hooks), and codify legal and organizational accountability before scaling agentic deployments.