Agentic AI: What Leaders Wish They Knew Sooner
Executive Take
Leaders deploying agentic AI should treat "human-in-the-loop" as a design problem, not a checkbox if approvers can't meaningfully engage, redesign which decisions require human review rather than adding more approval steps; and resist vendor labelling of "agent" as a proxy for capability maturity.
Executive Summary
At the 2026 MIT Sloan CIO Symposium, experts assessed agentic AI adoption. Thomas Davenport (Babson College/MIT) warned human oversight of AI agents is becoming performative, as approvers lack real time to engage. George Westerman (MIT Sloan) said most agents "are not really ready for prime time," with the term "agent" often overstating actual sophistication.
Why It Matters
Technology and AI leaders evaluating agentic AI rollouts need to know that oversight mechanisms widely marketed as safeguards may be functioning as rubber stamps, and that "agent" branding is outpacing actual capability both directly affect governance risk and vendor evaluation decisions.