AI Agents at Work: Efficiency Gains, Cognitive Pains

Agentic AI carries a compelling promise: systems that autonomously pursue goals, make decisions, and act—freeing humans from routine work. In practice, however, autonomy increasingly requires more oversight, not less. Inside organizations, AI agents quietly reshape behavior: meetings thin out, outputs are trusted without scrutiny, and decision-making becomes less transparent. What appears as efficiency can mask growing dependency, governance risks, and loss of overview. At the same time, generative AI is shifting work from execution to constant supervision—fueling cognitive overload, declining confidence, and what is now termed “AI Brain Fry.” This keynote explores the widening gap between promise and reality, where agents behave less like professionals and more like capable but unpredictable apprentices. The challenge is no longer building agents, but calibrating trust and redefining human–machine boundaries. How do we design systems that augment without eroding autonomy? And do individuals still have the right to opt out of AI-driven work—or has that choice already disappeared?