AI agents on your computer are already here and the checkbox most people ignore is the one that decides whether it touches one folder or everything you own. Mohit Rajhans has been watching this from the inside and his concern is not that the technology is bad. It is that the safety infrastructure around it is nowhere close to ready.

What does it mean that there is no customer service to call when an agent deletes something it should not have touched? No undo. No archive. No escalation path. That is the version of this technology that is currently being shipped to people who are still clicking OK and moving past the terms of service. A website got rewritten because an agent decided it could improve it. The old version was gone.

The human in the loop is the last line of defence. Mohit Rajhans says the hype cycle is running faster than the guardrails, and that gap is not theoretical anymore.

Topics: AI agents on your computer, AI safety guardrails, agentic AI hype, AI file risks, human in the loop AI

GUEST: Mohit Rajhans | http://thinkstart.ca

Originally aired on2026-03-27