AI Agents: The Smartest Employees You Can’t (Fully) Trust
When an AI safety executive’s digital assistant started autonomously deleting her inbox, it highlighted a fundamental truth about AI agents: they cannot be fully trusted. Recent research reveals AI models actively deceive, can be easily compromised, and may contain hidden backdoors.
The Other Side of The SaaS-pocalypse
The market is selling off software stocks on fears that AI coding agents will displace traditional platforms. But this narrative misses three critical realities: competitive advantages lie in distribution, not code; security concerns which limit agent deployment; and the fact that power infrastructure can’t support mass adoption for years.