Agentic AI accountability creates a genuine management puzzle
AI accountability isn't just about rules—it's about redesigning management for systems that move faster than human judgment.
I've been following the debate about autonomous AI oversight, and the results of the MIT Sloan panel capture something important. Sixty-nine percent of AI experts believe traditional management will not be effective for systems that make independent decisions and take actions on their own.
Speed is the practical problem. When an AI agent moving at superhuman pace makes a decision that causes harm, proving what happened becomes a nightmare. GitHub's chief legal officer nails it: today's workflows weren't built for AI operating at this speed, so we need clearer decision pathways and redesigned processes for tracing AI-driven decisions.
The "make the implicit explicit" recommendation hits home for me. With human employees, we rely on judgment and unspoken understanding of boundaries. Agentic AI systems need explicitly defined rules, threshold values, and escalation protocols—a fundamentally different management approach.
The quarter of experts who disagree point to complex systems like algorithmic trading and aviation that we already manage. But those systems don't autonomously generate new systems or adapt their decision-making in real time the way agentic AI can.
Product and legal teams need to build accountability structures into the design phase, rather than scrambling to address problems that surface later. How do you structure oversight when the system acts faster than humans can step in? That's what we're figuring out—redesigning management for systems that outpace human judgment.
Elizabeth M. Renieris, David Kiron, Steven Mills, and Anne Kleppe
